I finally found the time today to dig into oldsys-preseed and what I found wasn't pleasant. In fact, I may have solved the mystery why some people trying to install Debian on their NSLU2 cannot connect to the installer via SSH. oldsys-preseed is used on certain NAS devices to read the network configuration from the original firmware and these values are used to bring up SSH in debian-installer. If you used DHCP in the original firmware to obtain an IP address, debian-installer will do the same. If you have a static IP config, debian-installer will use it. In theory, it's all pretty simple, but users often run into troubles when a static IP config is used (as is the case on the NSLU2). While you only need an IP address, netmask and gateway to bring up SSH (and then you could configure the rest, such as DNS and hostname), debian-installer expects everything at the same time. I've seen many cases where users couldn't connect to the installer because they forgot to configure their DNS in the original firmware or some other values they didn't think were important. Anyway, right now oldsys-preseed will use a static configuration if the firmware is configured this way even if it's clear that netcfg won't like this configuration because of missing values. The IP address and DNS are always required, but the netmask and gateway can be figured out automatically. I wanted to add a sanity check that would choose DHCP in case the IP address or DNS were not specified. While working on this sanity check, I noticed two things: First, the "empty" value oldsys-preseed preseeds when there's no netmask or gateway is not accepted by netcfg and leads to an error message (which the user cannot see because SSH is not up yet). I'm pretty sure I tested this scenario when I wrote oldsys-preseed but it clearly doesn't work today and I don't see how it could ever have worked. This is pretty significant because having an empty gateway is a valid configuration (if hopefully not too common). Second, it seems that it's impossible to create a preseed file that tells netcfg that no gateway should be used. If you preseed an empty value, netcfg will simple use .1 as the gateway. I've fixed the first problem and proposed a patch to put in a special case to allow preseeding of no gateway. The sanity check that chooses DHCP when the network configuration is incomplete is also there now.

Reportedly, the USA is falling behind the rest of the world in bandwidth:

The 2008 median real-time download speed in the U.S. is a mere 2.3 megabits per second. This represents a gain of only 0.4 mbps over last year s median download speed. It compares to an average download speed in Japan of 63 mbps, the survey reveals. US also trails South Korea at 49 mbps, Finland at 21 mbps, France at 17 mbps, and Canada at 7.6 mbps, and the median upload speed was just 435 kilobits per second (kbps), far too slow for patient monitoring or to transmit large files such as medical records.

But don’t tell Chris Blizzard’s commenters. He writes about Comcast’s annoucement of a 250GB/month bandwidth cap, and gets an earful from commenters from Canada and Europe:

A boo hoo hoo. Major Canadian ISPs have had a limit of 60 GB for months, if not years. Oh wait probably the same way as most of the world manages on 10-20GB, for far more money than you re paying for $250. Not a lot of sympathy from this corner Yep, no sympathy from here either in Australia, with the only _independant_ ISP left, $280 AUD gets you 100GB. $50 with a major telco (the rest of the ISPs here) gets you 5GB. eg with my current ISP, a 8 MB line with a 300 GB monthly cap costs 20 GBP/month. A 8 MB line with unlimited bandwidth costs 160 GBP/month. Quite a difference! I pay the equivalent of $40 a month for 30GB, and extra GB on top are $3 each. That s with Plus Net (http://www.plus.net). I m in South Africa paying about $130 for a 10GB cap.

So who’s really better off? By my calculations, if a Canadian ISP provides 7.8 mb/s with a 60 GB cap, that’s about 17.5 hours per month of sustained maximum bandwidth before you’ve blown your limit. By contrast, an American ISP with 2.3 mb/s and a 250 GB cap gives you about 247 hours per month of sustained maximum bandwidth. Perhaps part of the answer is that only one country–Canada–shows up in the list of “faster countries” and in the comments section of Chris’s post. That could explain the apparent disconnect; maybe Great Britain and Australia are worse off than the USA, while Finland and Japan are better off. Still, this does bring the question to mind: which is better, raw speed, or the ability to actually use it without fear?

Rebuilding packages fast with distcc and ccache. I've found that I've neglected ecasound package for a long time. It's a pretty big C++ package, and one of the reasons I don't want to touch it recently is the time it takes to build the package. I played with distcc/ccache a little bit and found that it's pretty easy to set it up working with debuild thanks to those people who made it possible. debuild (or dpkg-buildpackage) now takes -j option for parallelism, and there are some options for ccache/distcc. I just crafted ~/.devscript, and ran debuild -j5, and the build went zooming fast. Cool.

As September is traditionally the "conference month" I'll be travelling all around Germany within the next weeks. So, inspired by Matthew, here's a short list of events I'll attend in September 2008. If anyone's up for a coffee, keysigning or something let me know!

• August 31st to September 4th 9th European Conference on Case-Based Reasoning, Trier, Germany
• September 5th to September 7th 3rd Congress on Solidary Economy Bremen, Bremen, Germany (Will probably arrive on Saturday)
• September 23rd to September 26th 31st edition of the Annual German Conference on Artificial Intelligence, Kaiserslautern, Germany (Will probably arrive on Monday)

As per the original press release:

In Quebec, access to public markets is the rule while contracts attribution without invitation to tender is the exception. A public market should be transparent, fair and most importantly, open to all. The solutions as well as the propositions must be evaluated objectively on known and accepted criteria. Furthermore, the regulation implies that public markets have to enhance the local economic development as well as the Quebec technologies. From February to June 2008, FACIL has noticed sales of proprietary software for more than 25 million dollars. These purchases were made for products offered by large multinational enterprises, with no regard to suppliers in Quebec.

Much to my surprise, news about this lawsuit have spread as far as Vietnam! Something tells me that this court case might have lasting repercussions...

In the mid 90’s I was part-owner of a small ISP. We had given out Trumpet Winsock [1] to a large number of customers and couldn’t convert them to anything else. Unfortunately a new release of the Linux kernel (from memory I think it was 2.0) happened to not work with Trumpet Winsock. Not wanting to stick to the old kernel I decided to install a Linux machine running a 1.2.x kernel for the sole purpose of proxying connections for the Winsock users. I had a 386 machine with 8M of RAM that was suitable for the purpose. At that time hard disks were moderately expensive, and the servers were stored in a hot place which tended to make drives die more rapidly than they might otherwise. So I didn’t want to use a hard disk for that purpose. I configured the machine to boot from a floppy disk (CD-ROM drives also weren’t cheap then) and use an NFS root filesystem. The problem was that it needed slightly more than 8M of RAM and swapping to NFS was not supported. My solution was to mount the floppy disk read-write and use a swap file on the floppy. The performance difference between floppy disks and hard disks was probably about a factor of 10 or 20 - but they were both glacially slow when compared to main memory. After running for about half an hour the machine achieved a state where about 400K of unused data was paged out and the floppy drive would then hardly ever be used. I had initially expected that the floppy disk would get a lot of use and wear out, I had prepared a few spare disks so that they could be swapped in case of read errors. But in about a year of service I don’t recall having a bad sector on a floppy (I replaced the floppy whenever I upgraded the kernel or rebooted for any other reason as a routine precaution). Does anyone have an anecdote to beat that?

• [1] http://en.wikipedia.org/wiki/Trumpet_Winsock

Share This

The Problem: A problem with virtual machines is the fact that one rogue DomU can destroy the performance of all the others by inappropriate resource use. CPU scheduling is designed to allow reasonable sharing of computational resources, it is unfortunately not well documented, the XenSource wiki currently doesn’t document the “credit” scheduler which is used in Debian/Etch and CentOS 5 [1]. One interesting fact is that CPU scheduling in Xen can have a significant effect on IO performance as demonstrated in the paper by Ludmila Cherkasova, Diwaker Gupta and Amin Vahdat [2]. But they only showed a factor of two performance difference (which while bad is not THAT bad). A more significant problem is managing virtual memory, when there is excessive paging performance can drop by a factor of 100 and even the most basic tasks become impossible. The design of Xen is that every DomU is allocated some physical RAM and has it’s own swap space. I have previously written about my experiments to optimise swap usage on Xen systems by using a tmpfs in the Dom0 [3]. The aim was to have every Xen DomU swap data out to a tmpfs so that if one DomU was paging heavily and the other DomUs were not paging then the paging might take place in the Dom0’s RAM and not hit disk. The experiments were not particularly successful but I would be interested in seeing further research in this area as there might be some potential to do some good. I have previously written about the issues related to swap space sizing on Linux [4]. My conclusion is that following the “twice RAM” myth will lead to systems becoming unusable due to excessive swapping in situations where they might otherwise be usable if the kernel killed some processes instead (naturally there are exceptions to my general rule due to different application memory use patterns - but I think that my general rule is a lot better than the “twice RAM” one). One thing that I didn’t consider at the time is the implications of this problem for Xen servers. If you have 10 physical machines and one starts paging excessively then you have one machine to reboot. If you have 10 Xen DomUs on a single host and one starts paging heavily then you end up with one machine that is unusable due to thrashing and nine machines that deliver very poor disk read performance - which might make them unusable too. Read performance can particularly suffer in a situation when one process or VM is writing heavily to disk due to the way that the disk queuing works, it’s not uncommon for an application to read dozens or hundreds of blocks from disk to satisfy a single trivial request from a user, if each of these block read requests has to wait for a large amount of data to be written out from the write-back cache then performance will suck badly (I have seen this in experiments on single disks and on Linux software RAID - but have not had the opportunity to do good tests on a hardware RAID array). Currently for Xen DomUs I am allocating swap spaces no larger than 512M, as anything larger than that is likely to cause excessive performance loss to the rest of the server if it is actually used. A Solution for Similar Problems: A well known optimisation technique of desktop systems is to use a separate disk for swap, in desktop machines people often use the old disk as swap after buying a new larger disk for main storage. The benefit of this is that swap use will not interfere with other disk use, for example the disk reads needed to run the “ps” and “kill” programs won’t be blocked by the memory hog that you want to kill. I believe that similar techniques can be applied to Xen servers and give even greater benefits. When a desktop machine starts paging excessively the user will eventually take a coffee break and let the machine recover, but when an Internet server such as a web server starts paging excessively the requests keep coming in and the number of active processes increases so it seems likely that using a different device for the swap will allow some processes to satisfy requests by reading data from disk while some other processes are waiting to be paged in. Applying it to Xen Servers: The first thing that needs to be considered for such a design is the importance of reliable swap. When it comes to low-end servers there is ongoing discussion about the relative merits of RAID-0 and RAID-1 for swap. The benefit of RAID-0 is performance (at least in perception - I can imagine some OS swapping algorithms that could potentially give better performance on RAID-1 and I am not aware of any research in this area). The benefit of RAID-1 is reliability. Now there are two issues in regard to reliability, one is continuity of service (EG being able to hot-swap a failed disk while the server is running), and the other is the absence of data loss. For some systems it may be acceptable to have a process SEGV (which I presume is the result if a page-in request fails) due to a dead disk (reserving the data loss protection of RAID for files). One issue related to this is the ability to regain control of a server after a problem. For example if the host OS of a machine had non-RAID swap then a disk failure could prevent a page-in of data related to sshd or some similar process and thus make it impossible to recover the machine without hardware access. But if the swap for a virtual machine was on a non-RAID disk and the host had RAID for it’s swap then the sysadmin could login to the host and reboot the DomU after creating a new swap space on a working disk. Now if you have a server with 8 or 12 disks (both of which seem to be reasonably common capacities of modern 2RU servers) and if you decide that RAID is not required for the swap space of DomUs then it would be possible to assign single disks for swap spaces for groups of virtual machines. So if one client had several virtual machines they could have them share the same single disk for the swap, so a thrashing server would only affect the performance of other VMs from the same client. One possible configuration would be a 12 disk server that has a four disk RAID-5 array for main storage and 8 single disks for swap. 8 CPU cores is common for a modern 2RU server, so it would be possible to lock 8 groups of DomUs so that they share CPUs and swap spaces. Another possibility would be to have four groups of DomUs where each group had a RAID-1 array for swap and two CPU cores. I am not sure of the aggregate performance impact of such a configuration, I suspect that a group of single disks would give better performance for swap than a single RAID array and that RAID-1 would outperform RAID-5. For a single DomU it seems most likely that using part of a large RAID array for swap space would give better performance. But the benefit in partitioning the server seems clear. An architecture where each DomU had it’s own dedicated disk for a swap space is something that I would consider a significant benefit if renting a Xen DomU. I would rather have the risk of down-time (which should be short with hot-swap disks and hardware monitoring) in the rare case of a disk failure than have bad performance regularly in the common situation of someone else’s DomU being overloaded. Failing that, having a separate RAID array for swap would be a significant benefit. If every process that isn’t being paged out could deliver full performance while one DomU was thrashing then it would be a significant benefit over the situation where any DomU can thrash and kill the file access performance of all other DomUs. A single RAID-1 array should handle all the swap space requirements for a small or medium size Xen server One thing that I have not tested is the operation of LVM when one disk goes bad. In the case of a disk with bad sectors it’s possible to move the LVs that are not affected to other disks and to remove the LV that was affected and re-create it after removing the bad disk. The case of a disk that is totally dead (IE the PV header can’t be read or written) might cause some additional complications.

• [1] http://wiki.xensource.com/xenwiki/Scheduling


• [2] http://www.cs.ucsd.edu/~dgupta/papers/per07-3sched-xen.pdf


• [3] http://etbe.coker.com.au/2008/05/22/xen-and-swap/


• [4] http://etbe.coker.com.au/2007/09/28/swap-space/

Share This

From a buildd log:

The following packages have unmet dependencies:
  apache2-threaded-dev: Depends: apache2.2-common (= 2.2.9-7) but it is not going to be installed
                        Depends: libaprutil1-dev but it is not going to be installed
E: Broken packages

Investigating with my set of perl scripts yields:

wouter@country:~/code/perl$ ./check-dep-waits
apache2-threaded-dev
dep-wait libmysqlclient15off (>> 5.0.51a)

Note the complete lack of any reference to mysql in the apt output. Also, in case you thought any of the above is bogus: apache2-threaded-dev depends on apache2.2-common which depends on libaprutil1 which depends on libmysqlclient15off. So there. I've long been hoping for someone to finally fix #325786, but that never happened. Now that I wrote that perl script that I linked to above, I don't think I still need it -- particularly because my script works on my laptop, too, rather than only on m68k machines. Which is good.

How to read the Freerunner's accelerometers This code has been take from moko_eightball by Jakob Westhoff: it just continuously prints the value of the three accelerometers.

The Linux kernel has a number of code sections which look at the apparent size of the machine and determine what would be the best size for buffers. For physical hardware this makes sense as the hardware doesn’t change at runtime. There are many situations where performance can be improved by using more memory for buffers, enabling large buffers for those situations when the machine has a lot of memory makes it convenient for the sysadmin. Virtual machines change things as the memory available to the kernel may change at run-time. For Xen the most common case is the Dom0 automatically shrinking when memory is taken by a DomU - but it also supports removing memory from a DomU via the xm mem-set command (the use of xm mem-set seems very rare). Now a server that is purchased for the purpose of running Xen will have a moderate amount of RAM. In recent times the smallest machine I’ve seen purchased for running Xen had 4G of RAM - and it has spare DIMM slots for another 4G if necessary. While a non-virtual server with 8G of RAM would be an unusually powerful machine dedicated for some demanding application, a Xen server with 8G or 16G of RAM is not excessively big, it’s merely got space for more DomU’s. For example one of my Xen servers has 8 CPU cores, 8G of RAM, and 14 DomUs. Each DomU has on average just over half a gig of RAM and half of a CPU core - not particularly big. In a default configuration the Dom0 will start by using all the RAM in the machine, which in this case meant that the buffer sizes were appropriate for a machine with 8G of RAM. Then as DomUs are started memory is removed from the Dom0 and these buffers become a problem. This ended up forcing a reboot of the machine by preventing Xen virtual network access to most of the DomUs. I was seeing many messages in the Dom0 kernel message log such as “xen_net: Memory squeeze in netback driver” and most DomUs were inaccessible from the Internet (I didn’t verify that all DomUs were partially or fully unavailable or test the back-end network as I was in a hurry to shut it down and reboot before too many customers complained). The solution to this is to have the Dom0 start by using a small amount of RAM. To do this I edited the GRUB configuration file and put “dom0_mem=256000” at the end of the Xen kernel line (that is the line starting with “kernel /xen.gz“). This gives the Dom0 kernel just under 256M of RAM from when it is first loaded and prevents allocation of bad buffer sizes, it’s the only solution to this network problem that a quick Google search (the kind you do when trying to fix a serious outage before your client notices (*)) could find. One thing to note is that my belief that it’s kernel buffer sizes that are at the root cause of this problem is based on my knowledge of how some of the buffers are allocated plus an observation of the symptoms. I don’t have a test machine with anything near 8G of RAM so I really can’t do anything more to track this down. There is another benefit to limiting the Dom0 memory, I have found that on smaller machines it’s impossible to reduce the Dom0 memory below a certain limit at run-time. In the past I’ve had problems in reducing the memory of a Dom0 below about 250M, while such a reduction is hardly desirable on a machine with 8G of RAM, when running an old P3 machine with 512M of RAM there are serious benefits to making Dom0 smaller than that. As a general rule I recommend having a limit on the memory of the Dom0 on all Xen servers. If you use the model of having no services running on the Dom0 there is no benefit in having much ram assigned to it. (*) Hiding problems from a client is a bad idea and is not something I recommend. But being able to fix a problem and then tell the client that it’s already fixed is much better than having them call you when you don’t know how long the fix will take.

Share This

I took a note of the various talks I attended at DebConf, with the aim of writing a few notes about each so that I could try to explain to my work mates exactly what I get out of going. However I think a lot of the benefit is about general face to face contact with fellow DDs and random conversations that happen. Having worked from home for over 3 years I'm finding it quite nice to be back in an office with people working on the same things as me; there are conversations that you will have over a quick cup of coffee or in the corridor that you wouldn't bother typing on IRC or email or picking up the phone for. Debian is a lot like a large organisation full of telecommuters and DebConf provides a vital opportunity for us to have those "minor" conversations that prove quite useful and to remind each other what we're actually like in person. I didn't end up writing a lot of notes during the talks so these are just snippets after the fact. If I'd been more organised I'd have done a write up on a daily basis rather than leaving it all until the week after. A lot of the below is fragmented thoughts, but the longer I leave it the less relevant it gets and the more chance I forget something. It'd have been up faster if I hadn't been fighting Movable Type to try and get some sort of headings. Sunday
Introduction, Marga The "Welcome to DebConf" bit. Marga got up and talked about how at DebConf4 (Brazil) she hadn't been a DD and now she was and how it had been a life changing event. She'd thought she was to blame for DebConf8 in Argentina, but trawling mail archives had shown it was originally Mart n's idea. She hadn't realised at the start quite how much work it would be. Steve's DPL talk, Steve McIntyre The DPL welcomes us to DebConf and talks about Debian's achievements, some of the results of his team survey (the teams largely say they need more manpower. Not really a huge surprise.). NM/User Survey BOF, Paul Wise Paul emailed debian-user, debian-newmaint and various other places (I think) asking about people's experiences as a Debian user or applicant to New Maint. His final results aren't yet published and I got the impression a disappointing number of our users had replied. The NM problems these days seem to be largely down to a lack of AMs, at least according to the stats on nm.debian.org. Meet SPI board - bdale, maulkin, jimmy, michael, others A chance to meet those of the SPI board who were present. The usual "If you're involved with Debian you should seriously consider joining SPI". Some discussion about the fact that although SPI was originally mostly Debian these days other projects are making more and more use of it - OFTC, PostgreSQL and Madwifi for example. Not a huge amount of new information for me. I've been an SPI member for some time and I think they perform a vital role for Debian. They've had issues in the past but it's really good to see them overcome and more projects on board. Monday
http://hp.com/go/debian, bdale Bdale talks about the history of Debian and HP. And makes the point that HP aren't doing any of this to be "nice"; their primary responsibility is to their shareholders. Not surprising, but it's good that even with that criteria at the fore front Debian ends up a good idea for a large multinational. Managing 666 packages, Mart n Ferrari Mostly centered around the pkg-perl experience and the tools/workflow they have to help manage things. Impressive web tool (DEHS) to help track packages with new upstreams, issues or pending uploads. Quality Assurance in Lenny + 1 I can't remember a lot of this, which is a bit bad. I remember a discussion about whether we should release with orphaned packages, or if they should be removed. The argument for not removing them seemed to revolve around a theory that if they weren't buggy then what was the harm, the argument for was that if they were unmaintained and not really used than how did we actually know they weren't buggy, and was it reasonable that they potentially held up library transitions etc? Bugs in large packages Don Armstrong and ways in which the BTS can help people with lots of bugs. This (thankfully) doesn't affect me, but it was interesting to hear Don talk about his work on the BTS and his various ideas. I was sorry I'd missed his talk on SOAP access to it; I shall have to find time to watch the video. Tuesday
Debian and Ubuntu, Mark Shuttleworth The usual interaction with Debian stuff, including how we interact with our upstreams. If a bug is raised with an Ubuntu package then it might well be appropriate to raise it with Debian and upstream as well, but each set of people should be able to use their own bug reporting systems rather than having to go to a foreign system. I don't think it was actually said, but I guess the idea is that launchpad is the thing that will do this gluing together. He also talked about Ubuntu as an "experiment" rather than a fork of Debian; something that has potentially different ideas about how to do things but that these ideas may lead to things that should be fed back into Debian. The point was made that even if you consider Ubuntu a fork they've forked 9 times now; it hasn't been a case of forking once and diverging wildly since thast point. I have a few thoughts on all the whole Debian/Ubuntu stuff which I should really try to form into something coherant and post. I think it's important to take Mark's talk in the context of bdale's "Shareholders come first" point. Derivatives round table Didn't stay for most of this. Ubuntu, Debian-EDU (which I hadn't realised was a derivative rather than a team within Debian, but apparently there are various things it was much easier for them just to patch now and try to work towards a generic solution later), someone from Extramadura and someone from the Munich government Linux stuff. I'm sure I've missed someone off. tdebs/i18n This was a useful BOF. It helped that we said no to the usual video team requirement that if you want to speak you had to wait for the microphone. This works fine for a normal talk, but really impedes the flow of conversation in something like a BOF. tdebs are something that have partly come from a need of Emdebian to reduce installed package size and partly from the desire of the i18n team to be able to update package translations without having to do a full binary NMU of a package. We had some of the FTP team, the release team, the i18n team and Emdebian there and I think some progress has been made. There'd been various conversations had before this meeting which I think helped people think about the issues and objections so that we had answers by the time we all sat down together - an excellent example of the use of DebConf. DebConf9 C ceres Various bits of discussion about next year's DebConf in Extramadura. Temperature is likely to be an issue for me as they reckon it could easily get up to 40C! I'll melt! Venue sounds good; some queries regarding late night hacklabs and net connections to same, but I'm sure the networking team will sort it all out as usual. ;) DebConf10 bids Venezela and New York were both pitching. Lots of people don't want to travel to the US or think they'll have problems doing so (I hadn't realised there was such an issue for people from South America). I suspect various people will have issues with Venezela as well; I heard expressions of doubt over their political stability expressed. However I'd be happy to go to either (and preferably both! DebConf10 + 12 perhaps?) Speaking of which, what ever happened to the Sarajevo bid? If they're reading, have you considered DebConf11? Wednesday
A day trip to an Argentinian ranch and lunch was an asido. Fantastic food and a pleasant afternoon spent sitting in the sun chatting to people. That evening was the formal, but thanks to having stayed up until after 4 on Tuesday I had to wimp out and go to bed fairly early. Not until I'd seen the DPL dance though. :) Thursday
Internationalization in Debian I forget most of this other than pretty graphs from Christian regarding what's translated and what other languages we might have; e.g. you can consider South America to be covered largely by Spanish and Portuguese, or you can consider it to not be covered very much at all by the local indigenous languages. dh-make-webapp: yeah right! Mainly what I took away from this was "It'll never happen thanks to the fact that every web app author uses a different system" and that more work needs to be done on the web apps best practices document. I don't really do web apps, but having them easily installable on my systems is a good thing so I hope some standardisation within Debian comes from this. Best practises in team-maintaining packages Various team members got up to talk about work flow in their teams. Not unsurprisingly different things seem to work for different people. Predictable PRNG in the Vulnerable Debian OpenSSL Package Lucian giving his Black Hat talk about the OpenSSL issues we faced earlier this year. Some queries about automated testing and that it should have found this, which originally I agreed with until I thought about it - you'd have to have run > 65536 tests and compared all the data to find this; something like the FIPS tests against a single run wouldn't have shown any issues. Lucian wanted feedback on ensuring he was presenting the Debian side of things acceptably as well. Keysigning Run by Don Armstrong. Nothing to really remark except I think key signing attendance is dropping off after things like the Helsinki experience. Also I've now managed to get my new RSA key signed by keys other than my own (and I'll try to get sorted out to sign keys myself soon). Friday
LessWatts Intel's drive towards reducing power usage. Odd to hear such things from someone who wasn't Matthew Garrett and I think I'd heard most of it before. Debian technical policy update Sorry Manoj, I should have been paying more attention. Debian Policy is something I think is important, but I'm not enough of a nitpicker to really feel I can contribute a whole lot. Virtualisation in Debian - Present and future This was depressing. What I took away was "We had 2 forms of virtualisation in etch (Xen + vserver) and we're supporting neither for Lenny". As someone running a Xen machine this is problematic. The host does have the hardware support that would allow kvm, but I'm worried about smooth migration (it's a colo box). There was some mention of Xenner, which isn't packaged for Debian. I took a look at doing so but there are some worrying items in the TODO list (like, oh, locking to enable reliable SMP) and I've spectacularly failed to get it compiling so far. I should have another go. Emdebian update Neil Williams about where Emdebian is and waved his Balloon3 about. Some interest from the Openmoko guys and discussion about why it might be better than OpenEmbedded. And equally what it couldn't do that OpenEmbedded can manage (mainly very small installs; a minimal Emdebian install is 24MB+ and with X that goes up to 75MB+ I believe). netconf I'd never actually seen Martin talk about netconf before. Am impressive flow of control diagram. It all sounds quite complex for little benefit at present, but if he pulls it off then I can see it being a much more flexible replacement for the likes of Network Manager. Saturday
Lenny - The Road To Release Neil played with putting the heads of the release team on random images. And tried to tell us all how to behave during the release process in terms of what should and shouldn't be uploaded etc. Most of it is just common sense, but apparently everything he talked about he'd experienced. Also pointed out we were supposed to release in 3 weeks and didn't even have the appropriate kernel in testing yet so it's probably not going to happen. Colour me surprised. Sustainable Computing Low power consumption/affordable computing. Interesting points about the OLPC being used in the middle of nowhere with no chance of an Internet connection being close by compared to neighbours in large cities where people can't necessarily afford top of the line computers or individual connections - in this sort of scenario mesh networking is really viable in terms of getting people connected up. Multi winner voting Manoj needs pretty diagrams, probably drawn by Martin Krafft. ;) Semi-interesting discussion about multi-winner voting (e.g. SPI board elections) and how the current Debian voting system is completely tied to Debian's infrastructure eg the LDAP setup. Manoj would like to make it more generic so it could be used by other groups as well as for different sorts of votes in Debian.

I'm back (for several days now) from Argentina and DebConf 8 to Portland. Too long a journey and too short a trip, as always. It was great to see everybody again as well as to meet a number of new folks from the local community. Here's hoping that DebConf 8 inspires more involvement in Debian from Argentina! It was my first time to this wonderful South American country, which made it even more melancholy to have to leave after such a short time and to be making this trip without Patty. Random observations about Argentina, for posterity:

• The 6-hour bus ride from Buenos Aires to Mar del Plata looks an awful lot like Iowa. Except for the occasional palm tree and the fact that it's 20 C in winter.
• Well, ok... Iowa also tends not to have quite so many cows. Or country grills (parrillas) right next to pastures...
• All the towns along the route are named after generals or colonels?
• All the police on the highway wear OSHA green vests.
• Following the German with the voluminous guidebook is an effective way to acquire generous portions of bife de chorizo.
• People from Bah a Blanca seem to have a somewhat more paranoid opinion of what constitutes a "dangerous neighborhood" than I do. :)
• "Buenos Aires" appears, I'm sad to say, to be a marketing lie. There was nothing bueno about the aire on the autopista on the ride to the airport. But the air in Mar del Plata was quite nice, and the people were the best!

As always it was a pleasure to take part in another annual DebConf, an event which is such a positive force for rejuvenating the Debian community. And BTW, Biella, NYC is not the only US city entertaining a bid for DebConf 10.

I got several emails in reply to my previous post on representing Sudoku problems as package dependencies. Several people asked about the algorithm that aptitude uses to solve these problems. A slightly out-of-date discussion package management and the aptitude algorithm can be found at http://people.debian.org/~dburrows/model.pdf. One question was interesting enough to deserve a follow-up blog post: Wolf T. referred to the page http://norvig.com/sudoku.html and asked how long it would take aptitude aptitude to solve the hard Sudoku problems on that page. He suspected that aptitude would do basic constraint propagation and then be forced to exhaustively search all the possible solutions to the problem. In brief: the answer is that it doesn't work by default, but not for the reason he suggested, and you can make it work with a few command-line arguments. In order to try this out you'll need a new version of the Sudoku reducer that can parse the problems on that Web page: debsudoku2.py. If I do a straight conversion of one of those problems and then run it through aptitude, it goes spinning off and never comes back. But aptitude is coded to avoid exhaustive searches; otherwise it couldn't solve dependency problems in reasonable amounts of time. In this particular situation, though, aptitude is clearly performing an exhaustive search; here's one series of status outputs:

open: 5124; closed: 4993; defer: 0; conflict: 374
Resolving dependencies...
open: 10090; closed: 9942; defer: 0; conflict: 398
Resolving dependencies...
open: 14726; closed: 14888; defer: 0; conflict: 440
Resolving dependencies...

Here open is the number of partial solutions that aptitude wants to examine, and closed is the number that it's already examined. aptitude isn't making progress here and it isn't finding solutions; in fact, it's creating, on average, about two new partial solutions for every one it examines. Why is aptitude doing this? Its resolver isn't a brute-force algorithm; in fact, it looks remarkably like the Sudoku solver discussed at http://norvig.com/sudoku.html. On each step, aptitude takes a partial solution, picks one dependency, and resolves it by installing or removing a package. This is pretty much equivalent to filling in one not-filled-in cell; in fact, in the conflicts Sudoku reduction, it's exactly equivalent. There is one difference, though, and it stems from a difference in the problem domains of Sudoku and package management:

Not all dependency resolutions are created equal.

In Sudoku, any solution to the puzzle will do. Not so in package management; solving a dependency problem by removing every package on the system will produce a formally correct answer and a very unhappy user. Even results that are not utterly disastrous could be annoying to the user, so the resolver needs to somehow aim for good answers. The page on Sudoku solvers suggests sticking to whatever value you placed in a cell, unless that doesn't lead to any solutions at all. This is technically known as a depth-first search. The problem with taking this approach when resolving package dependencies is that an apparently safe choice can end up requiring undesirable actions later on (like removing large swathes of the system). For this reason, aptitude tries to strike a balance between sticking to its current line of reasoning, and abandoning it when it seems to be turning out badly (this makes it a form of best-first search). On typical Debian archives, the package relationships are simple enough that it's OK if aptitude does this from time to time; it makes the search a little more thorough (thus costly), but the ability to back out of questionable branches is very valuable. But it looks like when aptitude is applied to Sudoku puzzles, it has trouble settling on a single series of choices. It keeps thinking that the solutions are getting worse, so it backs up and tries another branch, but they don't look any better over there, so it backs up and tries another branch, and so on. The option Aptitude::ProblemResolver::StepScore sets how hard aptitude will try to stick to the same line of reasoning. Its default value is 70, which is reasonable for package archives, but turning it up to a large value such as 10000 will cause aptitude to use a straight depth-first search, just like the Sudoku page recommends:

$ aptitude -s -o 'aptitude::problemresolver::stepscore=10000' -o 'aptitude::auto-install=false' install puzzle

Setting auto-install to false is a workaround for another odd corner case/bug that happens to bite this Sudoku problem: for technical reasons, doing this lets aptitude know that it should avoid solutions that cancel the installation of puzzle. Otherwise you have to sit there telling it no for a bunch of solutions you aren't interested in. Future versions of aptitude will probably provide a way to explicitly give hints to the resolver. With these options, aptitude was able to solve the hard Sudoku puzzles that I gave it in anywhere from fifteen seconds to a minute and 47 seconds. In contrast, the straight Sudoku solver does 30 problems in a second. Why is aptitude so slow? Based on skimming debug traces and looking at the run-time behavior, I conjecture that there are several reasons:

1. The aptitude search algorithm is very heavyweight. In the case where it takes over a minute, it only actually examines about 5,000 possible solutions. I don't know exactly where all this time is going, but I do know that the resolver does a lot of work on each step, for instance to try and detect dead ends that it hasn't reached yet. This is the conflicts counter that appears while the resolver is running. This pays off tremendously in terms of allowing aptitude to skip huge amounts of work (it may be why aptitude can solve Sudoku puzzles at all!), but it means that the resolver takes a long time for each potential solution it looks at.
2. The Sudoku page suggests taking the cell with the fewest number of possible entries as the choice. I distinctly remember writing code to do this in aptitude, but it looks like I threw it out at some point or never committed it. I don't see why it should be any worse than the current approach, though (which is to pick an arbitrary dependency); maybe it just got misplaced at some point. Implementing this behavior decreased the time required to solve one puzzle from a minute and 45 seconds to just under a minute; I've committed this change to the post-lenny branch.
3. The Sudoku resolver has the advantage of being able to very quickly identify illegal moves. aptitude's resolver can learn this information, but it takes time and it can't apply Sudoku-specific optimizations. For instance, the Sudoku solver knows exactly which values can be placed in a given cell at any time. aptitude has to actually consider each of the cells that's available as a possibility and prove to itself that, e.g., 4 and 6 can't be written in the same cell at the same time. And although it can learn this information, it takes time to do so and it takes time to see whether any of the past conflicts that it remembers are present in a given solution.

As I noted above, I've already checked in one change to the resolver's behavior. It might also be worth seeing what happens with a massive increase in StepScore, so I've pegged it at 10000 on my computer to see whether the results I get are noticeably worse. The third item above -- the fact that aptitude needs to do more work to understand which packages can't be installed -- is interesting. The Sudoku solver is very proactive about this: it keeps track at all times of which cell values are legal, and updates these sets every time it transitions. In contrast, aptitude constantly re-checks all the possible moves it could make to see which ones are currently legal. I think that the resolver's speed could be boosted a little by doing the following:

1. Cache the currently legal solutions of each dependency at the time it's inserted into the queue.
2. When a new search node is generated by resolving a dependency, drop the formerly legal moves that are knocked out (for instance, by installing a different version of the same package).
3. Come up with a quick way of re-testing new conflicts (the global list of dead-ends) so that search nodes that are already in the queue can be checked against the conflicts that were added since they were queued.

These changes won't make aptitude able to solve problems that it couldn't solve before, or at least not very many of them, but they should make the resolver faster on every problem it encounters. However, I don't think I want to actually implement them. It would be tricky to confirm that the resolver got faster on real dependency problems (since it's normally almost instantaneous), the changes would make the resolver code more complex and fragile (less readable, less maintainable), and there's a real chance of breaking something. Unless there are cases where the resolver is currently too slow on real package archives that don't involve an exponential blowup, I'm going to shelve the idea for now.

Apparently the LUG’s website is still off-line and there’s no idea when it will return, so let me use this site to help announce that the LUG’s regular meeting at the Knights Templar public house, near Isambard Walk, Bristol, will happen tomorrow Saturday 23 August from noon-ish until whenever people drift away. Software Freedom Day happens between this meeting and the next one. Is anyone interested in running a Bristol-area event and do you know of a suitable/available venue? I’m finding the SFD official site somewhat dysfunctional (I’m not in the Northern Caucasus, damnit), so I’ve not found a nearby team registered yet…

Docking the FreeRunner into the laptop Earlier in the day, I wrote:

So yes, my laptop can now be turned into a phone charger with networking, DNS and apt cache services. I shall look into hooking that script into dbus to have it run automatically when the phone is plugged and unplugged.

It's not dbus, it's udev, and I've managed to do it. It's an udev rule:

# cat /etc/udev/rules.d/z60_openmoko_net.rules
ACTION=="add", SUBSYSTEM=="net", ATTRS idVendor =="1457", ATTRS idProduct =="5122", RUN+="/root/bin/share-openmoko"
ACTION=="remove", SUBSYSTEM=="net", INTERFACE="usb0", RUN+="/root/bin/share-openmoko"

And a script:

# cat bin/share-openmoko 
#!/bin/sh
if [ "$ACTION" == "add" ]
then
    ACTION=start
fi
if [ "$ACTION" == "remove" ]
then
    ACTION=stop
fi
INTERFACE=$ INTERFACE:-"usb0" 
ACTION=$ ACTION:-"$1" 
case "$ACTION" in
    start)
        logger -t openmoko "Connected, setting up network"
        iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE -s 192.168.0.200/29
        ifconfig usb0 192.168.0.200 netmask 255.255.255.248
        /etc/init.d/dnsmasq start
        echo 1 > /proc/sys/net/ipv4/ip_forward
        ;;
    stop)
        logger -t openmoko "Disconnected, bringing down network"
        echo 0 > /proc/sys/net/ipv4/ip_forward
        /etc/init.d/dnsmasq stop
        iptables -t nat -F POSTROUTING
        ifconfig usb0 down
        ;;
esac

The script has extra cruft that makes it double as a udev script and as a init.d style script, just because I didn't feel like abandoning the init.d style interface. However, udev handles it perfectly, so there's probably no use at all for the start/stop part. This means that the interface and all supporting services will be brought up and down when the phone is connected/disconnected, and also when the phone is suspended/resumed. Of course, more fancy things can be plugged into the script, like syncing PIM info, file systems, turning on applets in panels, mounting phone file systems and whatnot.

Unpacking the new FreeRunner I got myself a FreeRunner. Here are some notes from the first few days: Available operating systems The FreeRunner comes with the Om 2007.2 distribution: it works for basic phone things, and it has an opkg package manager that you can use to install all sort of extra software. Only issue: the SMS application was rather unstable with my SIM card. I then tried Om 2008.8, which is a major new redesign, partly based on Qtopia. It's definitely more advanced on how it looks, with smooth animations all around, but I did not manage to get the GSM part to work, because I did not manage to get it to ask for a PIN. I then tried plain Qtopia and I got what looks and feels like a properly working mobile phone. The only issue that it has is that I did not manage to make it suspend, so battery life is much shorter than it could be. Apparently, this should improve as kernel 2.6.26 reaches the phone. I kept Qtopia installed in the phone flash as a "stable" phone system. And finally, Debian. Debian is based on the freesmartphone.org software stack, which is an attempt to create a UI agnostic DBUS frontend to the hardware that supports multiple applications running on top of it. It is a young project, but it can already drive a mobile phone in a useful way. It has a demo interface that is basically a showcase of what is implemented in the DBUS frontend, but does most of the basic things you need from a mobile phone, including taking and making phone calls. And then it has Debian behind, all of it. I need to buy a bigger microsd card. Tips and tricks Flashing things

1. apt-get install dfu-util
2. Start the Freerunner bios/bootloader, by holding down Power and then pressing AUX (standard bootloader), or by holding down AUX and then pressing Power (factory, unbrickable, read only fail safe bootloader).
3. dfu-util -l shows you a list of devices it can access. If you see more than once, you need to specify in all dfu-util commands which one you want, using -d USBID. In my case, I have to always use dfu-util -d 0x1d50:0x5119
4. To flash the kernel: dfu-util -a kernel -R -D /path/to/uImage
5. To flash the root file system: dfu-util -a rootfs -R -D rootfs_filename.jffs2
6. To flash the bootloader: dfu-util -a u-boot -R -D uboot_filename.bin
7. To flash the u-boot configuration: dfu-util -a u-boot_env -D env.new

You can also download all of these things from the FreeRunner by using -U instead of -D. Networking via USB All of the distributions I tried, by default configure the USB as a gadget with ethernet over usb. You can form a lan with it using the cdc_ether module, and you will find your phone preconfigured as 192.168.0.202 expecting to find a gateway at 192.168.0.200. I made myself this script to start and stop networking with the phone:

#!/bin/sh
case "$1" in
    start)
        iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE -s 192.168.0.200/29
        ifconfig usb0 192.168.0.200 netmask 255.255.255.248
        /etc/init.d/dnsmasq start
        echo 1 > /proc/sys/net/ipv4/ip_forward
        ;;
    stop)
        echo 0 > /proc/sys/net/ipv4/ip_forward
        /etc/init.d/dnsmasq stop
        iptables -t nat -F POSTROUTING
        ifconfig usb0 down
        ;;
esac

Besides doing masquerading, it also brings up dnsmasq so that the phone can always find a DNS together with the router. Another useful trick is to configure the phone to share the approx cache with the laptop:

deb http://192.168.0.200:9999/debian unstable main
deb http://192.168.0.200:9999/debian experimental main
deb http://pkg-fso.alioth.debian.org/debian unstable main

So yes, my laptop can now be turned into a phone charger with networking, DNS and apt cache services. I shall look into hooking that script into dbus to have it run automatically when the phone is plugged and unplugged. Changing the ringtone in Debian In case you don't like the default ringtone, this is how to change it:

1. vi /usr/share/python-support/fso-frameworkd/framework/subsystems/oeventd/receiver.py
2. look for def _play( self ):
3. change the codec in decoder = gst.element_factory_make, if needed. The list of available plugins is on the gstreamer website.
4. change the path in filesrc.set_property
5. /etc/init.d/fso-frameworkd restart and maybe /etc/init.d/zhone-session restart

Yes, the ringtone is currently hardcoded, but it does say it's a prototype after all. Or, if you prefer, it's fully configurable and the configuration can be found in /usr/share/python-support/fso-frameworkd. Configuring the bootloader You can connect to the u-boot bootloader via a serial terminal on /dev/ttyACM0. Type help and you will find that it can do a lot of things. The OpenMoko wiki has pages on the bootloader itself, its commands and the environment. The environment is the configuration of the bootloader, similar somehow to /boot/grub/menu.lst. Unlike grub, you can edit the environment from within the bootloader and then save it using the saveenv command. What I did:

1. A boot entry for Debian, with the kernel on an ext2 partition instead of fat: setenv menu_3 Boot from microSD (FAT+ext2): setenv bootargs \$ bootargs_base rootfstype=ext2 root=/dev/mmcblk0p2 rootdelay=5 \$ mtdparts ro\; mmcinit\; fatload mmc 1 0x32000000 \$ sd_image_name \; bootm 0x32000000
2. Stay in the bootloader until a choice has been made: setenv bootdelay -1
3. Don't power down the bootloader when idle: setenv boot_menu_timeout 99999
4. Always show the menu at boot: setenv stop_in_menu yes
5. saveenv

So now my phone dual boots, and I can choose if I want a more reliable phone now (Qtopia) or if I want to play with my future phone (Debian). Configuring ssh for two host keys on the same host Minor issue, but annoying: since both QTopia and Debian show up on 192.168.0.202, ssh will complain about changed host keys. Here is how to configure ssh to avoid the problem (in ~/.ssh/config):

Host debian
HostName 192.168.0.202
User root
HostKeyAlias debian
Host qtopia
HostName 192.168.0.202
User root
HostKeyAlias qtopia

I’ve ordered some USB charging leads for various portable devices, mainly because I’ve already got mains-USB and car-USB power adapters (they came with a bluetooth headset that charges from USB) and I agree with things like this:

“management of an individual charger for each device is becoming a pain in the backside. Since the USB power form specification is a standard, ANY charger with sufficient load capability should be able to charge ANY device.” (from USB As A Power Source)

but I’ve also heard that charging from USB is greener - or more precisely that because your computer’s power supply is running anyway, it uses less electricity to plug battery chargers onto the Universal Serial Bus than to plug in “wall-wart” DC adapters. But is this claim true? Claims from USB recharging product sellers concentrate on the obvious “less waste compared to disposable batteries” strength, although some smarter chargers cut power to charged devices - it’s not clear whether the mass-market USB battery chargers do this, but the reasons and the electronics don’t look that complicated. China (not famous for environmental protection yet) now requires USB-charging of mobile phones while a mobile phone industry body has recommended the same approach in general. However, there are concerns about whether Power over Ethernet is making networks less green, which might be true for USB if devices demand more high-power sockets and device makers don’t include intelligent control electronics (which people generally don’t want to pay for). Is USB charging greener than mains charging? I’m not sure and I haven’t found any great answers - can you? I guess in the worst case, I can try some tests with a power meter later…

I’ve just read an amusing series of blog posts about bad wiring [1]. I’ve seen my share of wiring horror in the past. There are some easy ways of minimising wiring problems which seem to never get implemented. The first thing to do is to have switches near computers. Having 48 port switches in a server room and wires going across the building causes mess and is difficult to manage. A desktop machine doesn’t need a dedicated Gig-E (or even 100baseT) connection to the network backbone. Cheap desktop switches installed on desks allow one cable to go to each group of desks (or two cables if you have separate networks for VOIP and data). If you have a large office area then a fast switch in the corner of the room connecting to desktop switches on the desks is a good way to reduce the cabling requirements. The only potential down-side is that some switches are noisy, the switches with big fans can be easily eliminated by a casual examination, but the ones that make whistling sounds from the PSU need to be tested first. The staff at your local electronics store should be very happy to open one item for inspection and plug it in if you are about to purchase a moderate number (they will usually do so even if you are buying a single item). A common objection to this is the perceived lack of reliability of desktop switches. One mitigating factor is that if a spare switch is available the people who work in the area can replace a broken switch. Another is that my observation is that misconfiguration on big expensive switches causes significantly more down-time than hardware failures on cheap switches ever could. A cheap switch that needs to be power-cycled once a month will cause little interruption to work, while a big expensive switch (which can only be configured by the “network experts” - not regular sysadmins such as me) can easily cause an hour of down-time for most of an office during peak hours. Finally the reliability of the cables themselves is also an issue, having two cables running to the local switch in every office can allow an easy replacement to fix a problem - it can be done without involving the IT department (who just make sure that both cables are connected to the switch in the server room). If there is exactly one cable running to each PC from the server room and one of the cables fails then someone’s PC will be offline for a while. In server rooms the typical size of a rack is 42RU (42 Rack Units). If using 1RU servers that means 42 Ethernet cables. A single switch can handle 48 Ethernet ports in a 1RU mount (for the more dense switches), others have 24 ports or less. So a single rack can handle 41 small servers and a switch with 48 ports (two ports to go to the upstream switch and five spare ports). If using 2RU servers a single rack could handle 20 servers and a 24port switch that has two connections to the upstream switch and two spare ports. Also it’s generally desirable to have at least two Ethernet connections to each server (public addresses and private addresses for connecting to databases and management). For 1RU servers you could have two 48 port switches and 40 servers in a rack. For 2RU servers you could have 20 servers and either two 24port switches or one 48port switch that supports VLANs (I prefer two switches - it’s more difficult to mess things up when there are two switches, if one switch fails you can login via the other switch to probe it, and it’s also cheaper). If the majority of Ethernet cables are terminated in the same rack it’s much harder for things to get messed up. Also it’s very important to leave some spare switch ports available as it’s a common occurrence for people to bring laptops into a server room to diagnose problems and you really don’t want them to unplug server A to diagnose a problem with server B… Switches should go in the middle of the rack. While it may look nicer to have the switch at the top or the bottom, that means that the server which is above or below it will have the cables for all the other switches going past it. Ideally the cables would go in neat cable runs at the side of the rack but in my experience they usually end up just dangling in front. If the patch cables are reasonably short and they only dangle across half the servers things won’t get too ugly (this is harm minimisation in server room design). The low end of network requirements is usually the home office. My approach to network design for my home office is quite different, I have no switches! I bought a bunch of dual-port Ethernet cards and now every machine that I own has at least two Ethernet ports (and some have as many as four). My main router and gateway has four ports which allows connections from all parts of my house. Then every desktop machine has at least two ports so that I can connect a laptop in any part of the house. This avoids the energy use of switches (I previously used a 24 port switch that drew 45W [2]), switches of course also make some noise and are an extra point of failure. While switches are more reliable than PCs, as I have to fix any PC that breaks anyway my overall network reliability is increased by not using switches. For connecting the machines in my home I mostly use bridging (only the Internet gateway acts as a router), I have STP enabled on all machines that have any risk of having their ports cross connected but disable it on some desktop machines with two ports (so that I can plug my EeePC in and quickly start work for small tasks).

• [1] http://www.darkroastedblend.com/2008/08/crazy-wiring-part-5.html


• [2] http://doc.coker.com.au/environment/computer-related-power-use/

Share This

This week has been a little hectic, as I've been struggling with testing different versions of the GNU/Linux kernel. Specifically I've been trying to solve a problem where a Phenom processor, when coupled with 8Gb, would kernel panic under heavy load. After testing various patches, kernel versions, and random things I believe the problem is fixed in the kernel version 2.6.27RC4 - however nothing in the changelog appears relevant, so I guess only time will tell. Now we need to solve the problem of Atom processors panicing when attempting to boot 64-bit kernels. That is still present in the 2.6.27RC4 kernel. (ObRandom: If there are any interested parties I can provide remote serial console access to such a system.) Finally I've also been playing with PAM, the plugabble authentication module. Again specific use-case here. At work we want to allow people to ssh to some systems (to access serial consoles, etc), and we wish their connections to be tested against our internal single-sign-on mechanism. That could have meant a whole new PAM module, which would do XML-RPC-fu. Instead it meant packaging libpam-external - which is a neat PAM module allowing you to specify a shellscript to validate users & passwords. (libpam-external is very similar to mod_authnz_external which is a similar pluggable Apache2 module) So, this week "kernel hacking", & "pam hacking". Does that make me a real developer now? ObQuote: Time Bandits

Just as a personal reminder for the next time I need that:

If there's a machine, for example a Xen host, with just one harddisk and you want to be somewhat safe in case of a disk failure, you can add a second disk and setup a RAID1 (mirroring). It's most simple to shutdown the machine and setup the RAID1 for the dom0. Afterwards you can start the domU VMs again when you're using LVM for their virtual disks. Following these steps will add RAID1 underneath your LVM setup:

0. /dev/sda5 is your current physical volume for your LVM, on the second (equally partitioned) disk /dev/sdb5 will be your starting point for your RAID1 (/dev/md3)
1. create a degraded RAID1 on /dev/sdb5:

# mdadm --create /dev/md3 --level=1 -f --raid-devices=1 /dev/sdb5
mdadm: array /dev/md3 started.
# cat /proc/mdstat
md3 : active raid1 sdb5[0]
135540288 blocks [1/1] [U]

2. prepare it for LVM and add it to your virtual group:

# pvcreate /dev/md3
# vgextend /dev/vg /dev/md3
Volume group "vg" successfully extended

3. move the extends from one disk to the other and be verbose to watch the progress:

# pvmove -v /dev/sda5
Finding volume group "vg"
Archiving volume group "vg" metadata (seqno 22).
Creating logical volume pvmove0
Moving 192 extents of logical volume vg/ts2-swap
Moving 1536 extents of logical volume vg/ts2-disk
[....]
Found volume group "vg"
Removing temporary pvmove LV
Writing out final volume group after pvmove
Creating volume group backup "/etc/lvm/backup/vg" (seqno 31)

4. remove the old partition from the virtual group:

# vgreduce /dev/vg /dev/sda5
Removed "/dev/sda5" from volume group "vg"

5. add the old partition to your raid and set it active:

# mdadm --add /dev/md3 /dev/sda5
mdadm: added /dev/sda5
# cat /proc/mdstat
md3 : active raid1 sda5[1](S) sdb5[0]
135540288 blocks [1/1] [U]
# mdadm --grow /dev/md3 --raid-devices=2
# cat /proc/mdstat
md3 : active raid1 sda5[2] sdb5[0]
135540288 blocks [2/1] [U_]
[>....................] recovery = 0.2% (284480/135540288) finish=39.6min speed=56896K/sec

6. let the RAID1 doing its sync and you're done.

Just as a small side note:
I don't know why the kernel don't want to boot when it get passed root=/dev/md2 as a kernel parameter. It works with root=/dev/sda3, though...