Older blog entries for jgg (starting at number 21)

So, I think I fount out what is causes saens to die, it isn't OOM as I thought, but it seems the process table is full. I installed watchdog+softdog and the machine sucessfully crashed and rebooted with log messages from watchdog indicating that it was unable to fork due to EAGAIN which kernel source seems to indicate no more task entries.

The question is why - which is why I wrote this single process pid logger thingy jiggy to watch out. Some buggy daemon is creating excessive processes, bets are its proftpd. I had forgotten that I ulimited the hell out of proftpd ever so long ago, so OOM it aint.

Why do we use this? Why do all FTPd suck? All I want is a bullet proof, fast anon-only ftpd that meets the DFSG. Sigh.

Of course with my luck it will be apache or some weird ass kernel bug. At least it reboots itself now.

Debian got a /48 of IPv6 addresses yesterday and I used them to create a couple tunnels and bring more of our machines onto the 6bone. For those that are not familliar with IPv6.. a /48 is large enough to hold about 64000 subnets if you use the usual configuration scheme. Each of those subnets can contain the largest switched ethernet network you can build (> 10000 hosts).

Debian Potato will have quasi usful IPv6 support in many packages, woody should be even better. The problem is getting upstream to accept patches. I wish more people writing socket/network applications would look at RFC2553 and visit the Man Page viewer at OpenBSD.org and look up getaddrinfo - that will enable their App for IPv6 pretty much right off. glibc since 2.1 has supported the functions, but there are no man pages since it is GNU.

At work today we got a new Printer/Copier/Fax. A Canon imageRunner 210. Apparently it costs quite a large amount of money new, something like 10k$ CDN or more. Really nice device, it has a built in 3G disk, postscript, 11" paper path and all digital copying with the usual bells and whistles. We even got the 12 tray finisher+duplexer with the stapler :> It is completely self supporting, you don't need a server to baby sit print jobs too it or anything!

Trouble is the Windows Drivers SUCK, the printer has so many capabilities they simply don't cover because they are stuck in the model of uni-directional communication with the printer :<

What I thought was kind of remarkable, they included a program that would download PDFs to the printer, it was supposed to take them internally as PDFs and rasterize them with the built in PostScript engine and the internal fonts. The docs say it is faster than printing from Windows and gave better results than the PCL driver.

Trouble is, Linux and this printer is worse :> Sure it speaks protocols that linux handles like SMB and LPR, but the totally balkanized way printing is handled on unix makes using the finisher exceedingly difficult. For TeX I need to add an arcane set of options to get duplexing and stapling, Netscape/Mozilla just doesn't have any hope and I imagine a2ps and enscript need another nasty set of options.

So I read the linux printing articles that have been floating around after the summit and it sounds very interesting, particularly the PPD programs - which will take care of the finisher, but it is still just as bad as the windows drivers WRT to font handling, programs pretty much have to download vector fonts with every job, even though I have 3G of storage that is quite capable of persistantly caching fonts! (only thing it is good for :P)

I hope the people involved with linux printing can get really good support for these expensive buisness class printers, it would make the MS office replacment' idea feasable...

It seems I did attend OLS and just got back. I think I saw everyone there that I knew, certainly a fair number of people.

The HP guys were especially nice, I think we will soon see a nice PA-RISC port for Linux.

It seems I will be attending OLS, leaving Tuesday. Hopefully through some magical means other Debian people will magically find me!

Otherwise, not much is going on.

Sooo.. I have begun my usual summer job. This time I am sitting in the RF chamber. a 6x12 foot room made entirely of plywood pressed between two layers of Mu metal.

It is grey. There are some 4 computers and 2 people in this poor little room, plus a pair of desks and some tables. Tsk.

This is what happens when you hire more employees than you have actual office space. Oh well. I spent the first little while ``acquiring'' a computer and monitor to use. This year I managed to get myself a PIII-600 and an Viewsonic 17" professional (the one with the short tube). Both are quite nice and very happy. I now have the best screen and best computer in the building by a wide margin ;>

The rest of the time I spend redoing networking cables that were wired by monkies not aware of the correct wiring standards and replacing faulty cards. Now most of the segments actually *don't* get thousands of framing errors.

Now that the US government has disabled the GPS dither I can use this differential GPS reciever here to compute my posistion down to about 10cm, which I also spent some time doing.. Too bad the GPS device isn't such a good time peice or I'd know what millisecond it was too.

Poor Debian killed, fixed, killed, fixed, killed, fixed poor auric, which is suffering from some kind of psychological disorder. It cannot decide if it is going to be stable, or unstable. Oddly, with SMP off it seems to be behave exceedingly well.. SMP tends to Ooops or OOM, or *something*. Distressing.

Saens returned and features Enhanced FTP Power. It is happily belting out 11mbit without even noticing right now. That should double up to 20 in a few hours. Gigabell is remarkably well connected internationally, it is just amazing. That alone should push up our usage quite high.

The list server got upgraded and demonstred it could do some 50 remote SMTP deliveries/sec (for about 20mins), which I think is reasonably impressive. That is some 4 million emails/day if it could sustain that rate.

I also converted master from qmail to exim. What a bloody nightmare that was. Ik. This does mean that new-maintainer only needs to add people to the LDAP directory, they do not need to do anything else like create home dirs and install .qmail files. This makes creating new accounts really fast and simple.

Having read Claudio's diary, I thought I'd give his modules a try.. Predictably, my poor player does indeed fail both. The S3M, I belive, fails because of an incorrect test for donness, jumping in and around the last order does that IIRC. Really simple fix. Neat song though. The MOD on the other hand.. Muse implements FastTracker stuff over ProTracker which of course screws it hard - they don't do jumps quite the same way, and I remember something about slides too being a tad different too.

It seems someone has been busy and fixed up MikMod this past year or two, the slink version wouldn't play either, but the potato version appears to get it OK. I wonder if anyone actually listens to modules anymore? At one point Muse was pretty damn good, it played everything, and at least for S3M, MTM, XM and IT/IT2 it was pretty much spot on. The rest of the formats (STM, WOW, FAR, 669, PTM and ULT) are OK for everything but the extremes. These days, I think, the new IT2s cannot be played (compression IIRC) and I imagine nobody writes anything but! I also never ported the GUS interface to linux. I really liked that bit of code too, sounded very sweet. Sad, bit rotted into obscurity :P

The Expell'O'Matic has served as the focal point for Debian's triannual flame fest. Those who haven't been around Debian too long find this pretty grusome. Roughly how it appears to go, is a good bunch of people select some arbitary issue and divide into factions and seek to destroy each other using bile, flames, acusations, facts (!) and what not. If the sides are unable to reach compromise and resolve the issues then a majority of all people participating will quit the project either because a) They feel people are unwilling to listen to them and so must quit or b) They feel directly responsible for the people who quit, and so must quit.

This form of brinkmanship appears to be completly unique to Debian, I haven't heard of any other projects who have the entire process quite down pat. Often other people see this and belive that Debian developers have a propensity to flamage and cannot be worked with.

Does anyone else remember the B5 where the Drazi divide up into factions based on colour of cloth? Think that. Anyhow, after uh, 3 days consensus seems to have formed, a result formed and the contestant has left the stage. Nobody is allowed to talk about what exactly happened publicly.

lilo leaving is rather disturbing, since it was my mention that brough him here.

5 days without an update Tsk Tsk. Oh well, I manged to write 3 finals, and I strongly belive I failed two of them - despite studying quite alot for one. I do not like transistors, I do not like transistors in OpAmp, Differential Amplifiers or in Cascode configurations. Yuk. Don't like FETS either. At least the last final appeared to go stunningly well, fancy that.

Being bored of studying I went to see the Fantasia 2000 IMAX before it goes away. Highly recommended, it was absolutely wonderfull. I'm so glad Disney decided to go the IMAX route. I also got to see the Dragon at the Theater breath fire. They seriously spent alot of money on that thing, it is huge and towers above the consession below this massive dome shaped roof. At some periodic interval spooky music starts going, lights focus on his head and after a bit it begins moving around, then finally the music peaks and this huge burst of real flame spits out of his mouth. The theater guys actually have a zone they keep clear during this process for safty reasons.

On that bored theme, I pulled down graphvis last night and produced.. Graphs. Fancy that. The key and the images. They are graphs of some of the package relations in Debian, people often wonder why APT makes odd automatic choices sometimes, well, these pictures explain it :P I build a tree for the whole project, but I can't graph it, I let lully grind at the data file for 88 mins and it didn't produce anything. I guess 22000 edges is too much. I highly recommend looking at base.png, it is quite interesting to look at the 76 base package relations.

saens is traveling to New York now, it is all upgraded and happy, The RAID can sustain a nice 16meg/sec semi-random seek which is great. I found there was a misconfiguration which was limiting speed in the old setup too. Oh well.

Murphy got upgraded to a new PII300 so it could fit in a 1U case. This has nicely managed to boost throuput by a few factors at least (procmail is slow). I don't know what the peak delivery rate of this new configuration is yet, but tests show that the round trip time is now averaging somewhere in the *20 second* region. Down from about 1-2 mins. linux-kernel is something like 2 hours! Crazy. Ben Collins also implemented a VERP front end for the list software which will even further help performance by removing the slow bounce processing step. This is significant because it means I am not going to write a MLM this summer, despite how interesting that might be, it is not needed.

Also I finished deploying LDAP on all the machines, which was then promptly used to indoctrinate bod into our little evil society. Offically now new-maintainer is open and people will be added at a steady rate. At 493 developers I wonder how high it will go.. I think this summer I am going to write some C++ <=> Python glue for my nice commandline/configuration library that APT uses and convert everything. Among other things.

Sadly, people are unfrieghtened by my mail proposal, and I implemented the configuration stuff for it as well. Just need to tidy up the LDAP data and we can move to multi-server central dispatch for @debian.org. Not that we need to or anything, but it simplifies a couple things that need doing.

I am moving, thus my cable modem has been returned and I am living the life of the disconnected modem user.

Joey, your Perl Scares me, but your self morphing 'objects' make me want to build a blanket fort in my living room and live out the rest of my years there.
Drow, the cabal owns you. lilo, IIRC I said something about why I cert'd you master on IRC.. Do you remember what it was? I forget.

We have another contestant for the Expell'O'Matic. Fear. I may very well have lots of hate mail by this time tomorrow.

I'm not sure why, but someone though it would be fun to play games with lists.d.o, 10000 emails later the problem was noticed and the crap erased. Still, who would do such a thing!? I also hate smartlist, horrid program wasn't designed to stand up to that type of abuse. At least Ben almost has VERP all worked out which will be a huge help. I've somewhat mentally assembled the bits needed to do a nice MLM and it isn't too terrible. Pity there are other things to do.

My first exam is soon, that will be nice because I don't have to study for it anymore..

I looked at the QNX hack, and I can't help but wonder 'huh?' I'm sure they have tested their crypt function against other crypt implementations and I know they have used shadow passwords for years - so what gives? If I had time I'd try it on our QNX 4 boxes and see what happens.

dhd only gets six Espies because his assembly was Alpha assembly, not IA32. IPv6 is also really cool, now that William is working on getting ftp.crc.ca fully on the 6bone I will have to finish my IPv6 implementation for APT's FTP method. I see also that the IPV6 tunnel on pandora is only configured to route 6bone test TLA traffic. I wonder what the correct configuration is supposed to be.