|
| Greg's diary |
| December 2004 |
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Wednesday, 1 December 2004 | Echunga | |
| Top of page | ||
| next day | ||
| last day |
More work on my documentation today. I've written something like 20 pages in the last few days, and the quality is “rough draft” at best. We need much more time for this.
In the evening to a celebration organized by the ICT council of South Australia (until very recently called the “IT council”; as of the time of writing, the web site still hasn't been updated), where to my surprise Ross Williams was awarded the Trevor Pearcey Young achiever award for an unspecified achievement. It took some explanation to find out that Trevor Pearcey was the designer of CSIRAC and had predicted, back in 1948,
It is not inconceivable that an automatic encyclopaedic service operated through the national teleprinter or telephone system, will one day exist
Ross picked up on this early anticipation of the world-wide web and defined a unit of foresight, the Pearcey, equal to 46 years, and discussed the effects of a decipearcy, a centipearcy, a millipearcy and even smaller. Interesting evening. At least we were able to postpone tomorrow's meeting yet again, this time until the others have read our documentation.
| Thursday, 2 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Now that I have the drafts of my documents out for review, got back to some real measurements on our software. That required a test machine. I still haven't completely recovered from my hardware catastrophes that started in June. I still have the dual processor zaphod working (slowly) as echunga, and so I had to use the machine designed to replace echunga for the tests, calling it beeble. teevee, the machine I bought to run the video setup in the lounge room, is currently called zaphod and running my /src file system. I didn't think before setting up /src as a UFS2 file system, so I can't move that to echunga until I upgrade echunga to FreeBSD release 5. What a pain.
Once I got beeble up and running, the fun wasn't over. Of course, the software version I had from last July didn't work: it died after a while with a claim that a magic number was wrong. I think it must have had a failed hack that I had put in at the time, but rather than debug it, decided to update to the latest version, since that's what I was aiming for anyway. That didn't work as well as it should have: for some reason, cvs kept dying on me:
U readfile.fw cvs [server aborted]: received broken pipe signal cvs [update aborted]: end of file from server (consult above messages if any)
With a couple of repetitions it was away, though, and after remembering that this build really needs gmake (when are we going to rename Makefile to GNUmakefile?), got it built and started.
While that went on—slowly—back to documentation. I had hoped to have less mail while people digested that, but that was not to be. There's also so much other stuff to tidy up. One change that I hadn't known about: we're dumping cvs in favour of subversion. Based on Michael Paddon's recent paper on the subject, I wonder if that's a wise move. It certainly requires me to think more.
| Friday, 3 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
I had intended to work on my documentation today, and indeed I did get some done, but not quite what I had planned. I normally write documents roughly like this:
I had supplied the documents in the middle of step (2), in the hope that the information would be of use. Instead I got a whole lot of feedback for step (6). The suggestions were for the most part valid, but that didn't help:
Nevertheless, ended up applying what I could.
| Saturday, 4 December 2004 | Echunga | Images for 4 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Somehow didn't get much done today. Bottling beer takes some time, but there should have been plenty left over to do other things. Spent some of it trying to install an “Instant Messenger” client, something I have been resisting based on the experience I've seen with it on Microsoft systems. The results were less than satisfactory. licq did this:
|
|
||||||||
To add insult to injury, it left my cursor changed to yellow, and I had to reset the xterm. Why do people make assumptions about the foreground and background colours of xterms (and then go and change them)? mutt does this too, and it's one of its most annoying “features”.
Why do ports install so badly? licq doesn't have a man page either, so gave up on that.
gaim looked more promising, but I couldn't work out how to configure it. It seems that “real” AIM clients register automatically with some central server, but for whatever reason, gaim didn't do that. Needed help on IRC to work out how to use it: it seems that you can register with AOL if you know how (either install an AOL client on your Microsoft box, or access this undocumented page). After doing that, things seem to be OK, though of course it would be too much to expect gaim to respect Emacs editing characters.
| Sunday, 5 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Spent most of the day getting AUUGN finished. The good news is that I managed to do it, but it's certainly very tiring. Hopefully I can find some more people to help me for the next quarter.
| Monday, 6 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
More work today; instead of continuing with my documentation, ended up handling still more mail messages. It's nice having some fresh ideas, but there's a point beyond which they just waste my time. Also some performance work to look at; it seems that I need to give that a whole lot more attention.
| Tuesday, 7 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Somehow I'm still flat out doing nothing very much. Received many mail messages requiring urgent attention, with the result that I didn't get through my mail until mid-afternoon, always a sign of overload. Then Ross called and we had a productive but long talk lasting the rest of the afternoon. I need more time.
| Wednesday, 8 December 2004 | Echunga | Images for 8 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
I have a new contender for “worst web site”: Virgin Blue, the Australian discount airline. Yvonne returned from Byron Bay today, and I was told flight number DJ 684. Tried the web site and selected “From: I'm not sure”, “To: Adelaide”, “Flight number: 684”, and got no responses. So I left out the flight number, and sure enough, no flight 684, in fact no flight that matched the information: no direct flights at all from that area (which has three or four airports). Finally tried each airport in turn and found that entering “From Gold Coast” did the trick. What a crock!
To make matters worse, had a minor accident on the way to the airport: going down Greenhill Road in the left-hand lane, hit another car with the tip of my door mirror, knocking out his door mirror. Interestingly enough, two cars were damaged, both blaming me. The first one definitely was my fault, but I have my grave doubts about the second one, which appears to have been damaged from the opposite direction. In addition, he was parked at least 15 cm further towards the pavement: if I had hit both cars, I would have run into the car parked in front. In any case, I only heard one bang. I suppose it'll be up to the insurers to decide.
We've moved to subversion now, and my tests require me to check out the latest version, so I had to install it. Not easy: something is broken in the automake port, and it will no longer build over NFS:
That meant finding the configuration log, of course, so it wasn't as simple to find as it looks. Looked at this autom4te program (why does the name irritate me?) and found it was written in perl, a language I don't use, and there was no obvious (to me) way to find out where it was failing. Gave up and installed the ports tree locally and installed like that. After installation, of course, the usual problems:
=== root@wantadilla (/dev/ttyp0) /usr/ports 435 -> man -k subversion
When will people finally get round to printing a summary of what documentation is available at the end of the port install?
| Thursday, 9 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
I still have so many things to do that I'm getting nothing done. Finally managed to find some subversion documentation. As yesterday's printout shows, there was some documentation in /usr/local/share/doc/subversion, but it's as good as useless. The only thing of use is the file README, which states:
The main documentation is the Subversion Book, written in
DocBook XML, which lives in the doc/ tree. If you wish to build
the documentation from source, read doc/book/README. Otherwise,
an on-line version of the book can be found at
http://svnbook.red-bean.com.
Sigh What happened to UNIX concepts of on-line documentation? In fact, there are man pages (at least svn(1) and svnadmin(1) (and not, as you'd expect, svnadmin(8)), but README doesn't mention them, maybe out of shame: they're as good as useless. And of course the URL above is incorrect: you've still got to go looking for the HTML and PDF versions of the documentation.
That documentation isn't too bad, and with a couple of hiccoughs managed to check out a source tree. By that time it was time to go to town to a meeting of the ICT council board, which went on for a record 210 minutes. Hopefully we'll be able to get the time down, though people are unwilling to defer more discussion to email. There's a general aversion to email on the board, probably because most people use Microsoft and blame Microsoft's failings on email.
After that to Dale Long's place to pick up yet another couple of Intel SMP machines, one of them a 4 way Compaq ProLiant 6500R. Not the newest of machines, but presumably the SMP issues will be the same. Including a couple of dubious ones, I now have 7 SMP motherboards.
| Friday, 10 December 2004 | Echunga | Images for 10 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Finally found time to look at the performance issues that I discovered last week. In the meantime, Tim Stoakes had established that the problem is in malloc(). Doubtless the there are too many calls to malloc(), but that shouldn't make things as slow as it does. After some investigation, discovered that it was related to the fact that the program uses pthreads. Apparently this slows down FreeBSD malloc() much more than it should.
With a bit of investigation found an alternative malloc() (in fact, the base for the GNU libc implementation, presumably what Linux is using). In the process of installing it, found a bug in the FreeBSD pthreads implementation, for which a fix had been posted six months ago, so committed that. After that, confirmed that it works much better: the test run that had run for 180 seconds on Monday now runs in 40. Considering that the program is presumably still spending considerable time in malloc, that's a remarkable performance issue for FreeBSD malloc(). Now to make a port of ptmalloc.
The issue isn't done with that, of course. FreeBSD is not the only system that is liable to have problems with malloc() in a threaded environment: there's evidence that Sun could have problems too.
In the evening playing around with one of the SMP boxes I got yesterday. It's interesting that Linux doesn't recognize the Compaq array controller (2DH), but FreeBSD does. Installed FreeBSD 4.10 on it with no difficulty (5.1 wasn't as happy and panicked during the probes).
| Saturday, 11 December 2004 | Echunga | Images for 11 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Why is it that so many web sites are so bad? I have a couple of cron jobs that download TV programme data from web sites and send it to me. Today the ebroadcast web site changed thir layout, and I no longer got the categorized pages. So I went off looking at the individual channel web pages.
We have five free-to-air channels in our area: three commercial stations (“Seven”, NineMSN (yes, Microsoft) and “Ten”) and two public channels (ABC and SBS). Of these, only SBS has an acceptable programme on the web site. ABC used to, but it would appear that they found it too easy to use (oops, sorry, that's probably “old-fashioned”), and they changed it to make it more like Ten: only a title in the overview, and for the rest you need to follow a link. I wonder how long it's going to take for people to realise that this doesn't make it easy to read. And of course they have to be “clever”, which translates to broken layouts. This one is from http://abc.net.au/tv/guide/ of 12 December 2004:
|
|
||||||||
Yes, part of the issue here is that I use a large font. That's not because I'm going blind; I display the text on a 2048x1536 monitor (see my hardware setup), and it's necessary to make it legible. This shouldn't be an issue, since HTML is supposed to make no assumptions about the display. But these sites do, and the assumptions are invalid.
This programme also has an excellent example of why it is useless. Looking at the following item (which is complete; I haven't chopped off anything), there is no information whatsoever about the content. I wouldn't normally look more closely, especially since “contemporary life” programmes are usually pretty boring. The only discrepancy is a travel advisory warning, which seems completely superfluous at this point, since nobody reading it would otherwise know that travel might be involved, let alone to where.
|
|
||||||||
I did follow this link (in case you're wondering: the text “Pilot guides” is the link) and found:
|
|
||||||||
This does look interesting. But why are they hiding it where most people will never see it? It's also interesting to note that the programme compilers seem to think that RECOMMEND TO A FRIEND is the most important part. How can I recommend anything about this programme? Also, what does world history have to do with contemporary life?
Some time ago I wrote to ABC complaining about the changes, and pointing out the breakage. I got no reply, which I suppose is normal nowadays. But why?
“Seven” and “NineMSN” supply programmes for all channels, not only their own content. Unfortunately, it's even more superficial. I can't understand why they don't give as much detail as possible, and why they don't seem to care if their content is obviously broken. Surely their advertisers care whether people watch the programmes or not.
In the afternoon, played around a bit with the other Compaq machine that Dale Long gave me, the 6500R. To my surprise, it just worked; at least, I was able to install FreeBSD 4.10 on it with no preparations whatsoever. Getting 6-CURRENT running is a different matter, but that seems to be more of a general issue.
| Sunday, 12 December 2004 | Echunga | Images for 12 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Yvonne left for a training course in Victoria today, and Yana went off to stay the night in Adelaide, leaving me alone here for the first time in a long time. Spent some time working on my temperature control software: I had had a problem that it can only display on one system, and that a network problem would cause the software to hang (and thus the temperatures to drift). Decided that a network queryable interface would be good, and set to doing it. I can now do:
=== grog@freefall (/dev/ttyp7) ~ 54 -> telnet brewer.lemis.com 4135
Trying 192.109.197.147...
Connected to brewer.lemis.com.
Escape character is '^]'.
Time Brew Brew Base Ambient Goal Offset Room
41 42
14:21:24 0.81 0.56 0.69 1.37 0.50 0.19 21.06
Status: Cooling
Connection closed by foreign host.
To do this, I also had to move brewer to a globally visible address, not as easy as it should have been. There's still something funny with my DNS setup.
The application should really use UDP, but the ability to query via telnet seemed valuable enough to use TCP instead.
In the afternoon, spent far too much time trying to write a little client to do this repeatedly. Somehow DNS lookups are just far too complicated, and getaddrinfo does nothing to simplify it.
In the evening to the airport to pick up my father, who is looking a lot frailer than last time I saw him at the beginning of the year. I don't suppose a flight from London with British Airways made things any better.
| Monday, 13 December 2004 | Echunga | Images for 13 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Somehow another day where I didn't get round to doing what I had intended. I had more or less finished a port of ptmalloc on Friday, but by this evening I had still not finished. The usual long mail exchanges seem to have been the main reason, and also installing FreeBSD on the Compaq 6500R (four processor). To my surprise, after finding the FreeBSD 5.2.1 CD-ROMs, I was able to start a perfectly normal installation with no special preparations. Then I started the upgrade to 6-CURRENT, which is related to my incomplete notes on upgrading a system, but of course I needed to do it differently. It's nice to have a start, though, but getting the documentation right seems to be a particularly tiring effort.
| Tuesday, 14 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Finally got round to committing my ptmalloc port today, and was promptly informed that it didn't run on FreeBSD release 4. Well, it does, but the tests don't compile: it's missing libpthread.so. Gave up on that and decided not to build the tests by default.
Also got FreeBSD 6.0 installed on quartet, and was able to do some reasonable testing, including the original program that had the malloc performance issues. How about that! It's not documented, but FreeBSD -CURRENT has the j option enabled by default. After disabling it, FreeBSD malloc significantly outperformed ptmalloc:
FreeBSD malloc default 939.37 real 672.82 user 253.10 sys ptmalloc 264.35 real 113.59 user 126.34 sys FreeBSD malloc, no debug 157.10 real 93.68 user 52.59 sys
I wish I had known that before I started. And wouldn't it be nice to be able to find out what the current defaults are?
| Wednesday, 15 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Has it really been over a week that I've been looking at this performance stuff? As usual that means that I left other things hanging, and I spent most of today catching up with them. Also started updating duet to FreeBSD 4-STABLE. It's been a long time since I last did any work on release 4, and it shows. At least by the evening I had things relatively well sown up; now I can move on.
| Thursday, 16 December 2004 | Echunga | Images for 16 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Spent most of the morning on the phone to Ross. Once again, we're changing direction, and we spent a couple of hours discussing what to do.
The range hoods that I bought last month have finally arrived, and this afternoon I went to pick them up. I should write a complete page just about the issues involved in getting something shipped from one country to another. Today involved a long report at the Customs Department, which is obviously not set up to handle this sort of thing efficiently. They were friendly enough, but a run-of-the-mill declaration took over 30 minutes.
Then to AQIS to confirm that there was probably no organic matter in the shipment. They were at 8 Butler St, and the next people I had to visit were the shippers—at 8 Butler Drive. Unfortunately, this was many kilometres away, and after paying that, I had to return to where I started, to Sideloader Shipping in Evans St. Found a way to Evans St. avoiding the big traffic, but had my difficulty finding the street: no street sign. Finally I found it and discovered that there was no Sideloader. After enquiring, discovered that Evans St. is only a fraction of its former self, blocked off over most of its length, and a dead end. No, make that two dead ends. I was in the wrong one.
With instructions and a more careful look at the street directory, headed off to the other remnant of Evans St., and missed it again. That wasn't difficult:
The second time round we found the name of the street, written on a lamppost:
|
|||||||||||||||||||||||
We weren't done, though: at the end of the road was a building, but it wasn't Sideloader:
|
|||||||||||||||||||||||
Sideloader itself was carefully hidden behind a barricade of containers. This view is about the only place where you can recognize anything at all, and to get there you have to drive to the other end of the compound:
|
|||||||||||||||||||||||
Still, got our goods relatively easily, despite having been given the wrong delivery order, and off back home, stopping off at Grumpys on the way. Andrew Schultz (“Grumpy One”) was showing why he got the name. It seems that the Mount Barker Council doesn't want to honour their liquor license, and Danny from the Mount Barker Courier (who visited me last year for the Telstra problems) was there taking photos of some unhappy looking people, myself included. Didn't get my malt as a result.
The article was published on 22 December
| Friday, 17 December 2004 | Echunga | Images for 17 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Another day spent mainly on the phone. I had barely got through my mail this morning when Ross called me, and we spoke for 2 hours. Once again a change of emphasis. As a result of that, spent a good part of the afternoon on the phone with Alan Kennington.
In the meantime set up to do some testing, but first decided to check what was on the disk I was going to use and discovered a bug in my mklinks program, so that needs to be clarified first.
| Saturday, 18 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Another day spent brewing. This just takes too long. Admittedly, it's not only brewing: I get a chance to do a little in between, but it means I can't do anything else. Spent most of the intervening time thinking about how to make it less time-consuming. Most of the ideas still seem to be related to making larger batches.
| Sunday, 19 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Attempted to relax today, without too much success. Spent some time looking at my temperature control software. Managed to get a TCP client to talk to the software, but didn't finish. It's really a little silly to use TCP to get the information; probably I should be using UDP and have the server handle both TCP (for telnet connections) and UDP (for repetitive stuff). Also got some request logging working:
Dec 19 12:11:16 brewer tempcontrol: Query from 12-202-208-105.client.insightBB.com (12.202.208.105) Dec 19 17:23:11 brewer tempcontrol: Query from 220-245-154-231-vic-pppoe.tpgi.com.au (220.245.154.231)
Also did some work on getting the plots working better, with some eventual success. I had two copies of the documentation, one of them done with groff, which was a mistake: the build target is broken, and the result bears little structural resemblance to the original. Finally reformatted the documentation with LaTeX and threw out both old copies (and, it eventuated, the index, which is gone from the new version). Found what I wanted to know, though: to format dates specified as UNIX seconds-since-the-Epoch timestamps, you write:
set xdata time
set timefmt "%s"
The %s is not documented in other places I found. It's also still not clear how to change this to local time, so currently my plots are showing UTC, but it's a huge improvement on earlier attempts:
| Monday, 20 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
More performance testing today, using quartet, my “new” four processor machine:
CPU: Pentium Pro (200.00-MHz 686-class CPU) Origin = "GenuineIntel" Id = 0x619 Stepping = 9 Features=0xfbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV> real memory = 268435456 (256 MB) avail memory = 253308928 (241 MB) FreeBSD/SMP: Multiprocessor System Detected: 4 CPUs cpu0 (BSP): APIC ID: 3 cpu1 (AP): APIC ID: 0 cpu2 (AP): APIC ID: 1 cpu3 (AP): APIC ID: 2
It's a good illustration of how things have progressed in the last five years. The application is CPU-bound and (currently) tied to a single processor, so instead of running for an hour or so, it ran overnight. I was using a couple of Vinum arrays, and since this machine is running FreeBSD 6-CURRENT, it was gvinum. It was quite gratifying to connect the disk trays to the machine, run a camcontrol rescan and see:
GEOM_VINUM: subdisk raid.p0.s0 is up GEOM_VINUM: subdisk raid.p0.s1 is up GEOM_VINUM: subdisk raid.p0.s2 is up GEOM_VINUM: subdisk raid.p0.s3 is up GEOM_VINUM: subdisk raid.p0.s4 is up GEOM_VINUM: subdisk raid.p0.s5 is up GEOM_VINUM: subdisk raid.p0.s6 is up GEOM_VINUM: subdisk blockpool.p0.s6 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s5 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s4 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s3 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s2 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s1 is up GEOM_VINUM: plex blockpool.p0 is down GEOM_VINUM: subdisk blockpool.p0.s0 is up GEOM_VINUM: plex blockpool.p0 is up GEOM_VINUM: subdisk raid.p0.s6 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s5 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s4 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s3 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s2 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s1 is up GEOM_VINUM: plex raid.p0 is down GEOM_VINUM: subdisk raid.p0.s0 is up GEOM_VINUM: plex raid.p0 is up
It would be nice to get rid of the shouting, of course, and some of those messages are spurious. But it's nice to see things Just Happen.
In the meantime cleaning up my other 160 GB disk, which will hopefully work a bit faster. Discovered a number of problems in my mklinks program, and fixed at least one, but was still left with about 10 GB on the file system, apparently mainly empty directories: I had forgotten to write that code. More for tomorrow.
| Tuesday, 21 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
More test work, this time with my program, and got far enough to discover (and subseqeuently fix) a 32 bit overflow. Pity it took so long to get there.
Also looking at mklinks, which is still not really complete. There are some interesting interactions between symlinks and deletion which could potentially end up with replacing a file with a broken symlink. Care is needed.
| Wednesday, 22 December 2004 | Echunga | Images for 22 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
More testing today; how slow this system is. I have yet another bug in my program, and I spent all day testing it (mainly waiting for the bug to show up). In the meantime worked on removing duplicate directories from my other disk with mklinks, which I improved considerably in the process, but I fear it's still not ready for release.
Today's Mount Barker Courier included an article about Grumpys, including one of the photos taken last week. It'll be interesting to see what happens. One of the objections quoted was:
Six people objected to the expansion on the grounds of traffic, noise and flood plain issues.
Grumpys' is directly on the freeway exit. It's difficult to see how it can compete, either with noise or with traffic, with that. And it's completely unclear to me what the flood plain has to do with anything, beyond the fact that the Mount Barker Council has been remarkably reticent to fix the problem. It almost looks like they're trying to get their own back for complaints about their own performance.
Yana has established a need for financial software, and I suggested GnuCash, possibly a little hastily. It's yet another GNOME application, and it pulled in 50 other ports, totalling nearly 600 MB of build space, and took hours to build. Where did UNIX simplicity go? In the end I started it (which took 18 seconds of CPU time on an Athlon 3200XP+) and it opened two of the tiniest windows I've ever seen:
|
|
||||||||
On a 2048x1536 screen it was barely recognizable, and the tiny fonts (about 4 pt) make it almost illegible. Spent some time searching through the documentation on a low-resolution screen (after first enlarging it to a reasonable size, a detail it refused to remember), and came across the following description of the Preferences menus:
4.2.11. Advanced: This section contains the following preferences (this screen is hidden by default); * Save Window Geometry:
All attempts to rectify this default failed. What a crock! It doesn't seem to be possible to set the font sizes at all; if so, they've hidden it very well.
In the evening, it looked like our mare “Narrawin's First Lady” (whom we call simply “Lady”) was going to foal, and spent most of the evening going out and checking. Nothing was forthcoming by the time I went to bed.
| Thursday, 23 December 2004 | Echunga | Images for 23 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Testing continued today, along with another flurry of mail messages. After running one series of tests, noted particularly the summary:
Started at 23-Dec-2004 15:36.09
Ended at 23-Dec-2004 15:59.44
Running time was 1415 seconds.
Total memory malloced was 79370K.
1419.87 real 707.31 user 467.35 sys
84828 maximum resident set size
...
1132267 signals received
47017 voluntary context switches
124192 involuntary context switches
Those signals looked interesting; nearly 1000 a second. Run ktrace and discovered:
9742 mystery PSIG SIGPROF caught handler=0x2816a3ec mask=0x0 code=0x0 9742 mystery Events dropped. 9742 mystery PSIG SIGPROF caught handler=0x2816a3ec mask=0x0 code=0x0 9742 mystery PSIG SIGPROF caught handler=0x2816a3ec mask=0x0 code=0x0 9742 mystery Events dropped. 9742 mystery PSIG SIGPROF caught handler=0x2816a3ec mask=0x0 code=0x0
In other words, it looks like I got at least in part a profiling executable. What a waste of time!
In the evening, Lady finally gave birth to her foal, a colt. Photos to follow.
| Friday, 24 December 2004 | Echunga | Images for 24 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Off into town ridiculously early to a meeting, the first I've been to in the Shell House offices since October. In the meantime we've hired a number of new people, and this was the first chance to talk to others who are working on the project.
After that off to the botanical gardens for lunch. Somehow I've lived here for nearly 8 years without being there. Quite a nice place, and we took our time over lunch, not leaving until after 3 pm.
| Saturday, 25 December 2004 | Echunga | Images for 25 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Christmas has been the same routine for the last 8 years now: see the family for (just about) the only time in the year, and eat lots of food. Today was the same, One amusing incident: I was talking to Marco about something, and he came up with a quote (in Italian) that roughly ran “Blessed are the one-eyed, for they shall rule the blind”. Checking back later, I found an amusing photo of him and Karen taken four years ago in almost exactly the same spot, with the title “In the land of the blind, the one-eyed man is king”.
|
|||||||||||||||||||||||
As usual, took lots of photos. Removed the dart from my ear:
|
|||||||||||||||||||||||
Back home and tried to digest, with only moderate success.
| Sunday, 26 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Another quiet day, as befits the second day of Christmas. I had intended not to do much with computers, but ended up spending a fair amount of time not only processing yesterday's photos, but also tidying up some of my older photos. There's a whole lot of stuff there that needs tidying up.
Also spent some time looking at my somewhat neglected computer video project, and finally got teevee back in its originally intended configuration (the disk had been running on an old dual processor machine while its own hardware had been simulating zaphod). In the process discovered that the infrared support promised for many motherboards is more virtual than real: the motherboard is an MSI K2 Combo, which boasts:
- 1 IrDA connector for SIR/ASKIR/HPSIR
After some searching, I found a page in the manual that said the same thing. But the list of connectors doesn't show any such connector; neither did an examination of the board. In any case, it looks as if IRDA (for which documentation was also singularly lacking in the FreeBSD manual) is not up to the task of understanding a TV remote control. Daniel O'Connor has apparently been dabbling in this stuff, so it looks as if he, Chris Yeoh and I will have a talk about it at the hackers' barbecue planned for next weekend. In the meantime got some potential feeds which I've put in their own section on the multimedia page.
| Monday, 27 December 2004 | Echunga | Images for 27 December 2004 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Intended to take it a bit easier today, and really did manage to finish playing with computers before lunch. Went out to look the sprinkler system. Circuit 4 was looking a little dry, and it proved that it hadn't been sprinkling at all: the wiring to the solenoid had been damaged.
While investigating that, however, the control unit died. After some cursing and swearing, discovered that the on-board power supply had burnt out:
|
|||||||||||||||||||||||
After some consideration (and further photos) decided that the thing wasn't worth repairing: I already had all components to replace it with a computer, though the relay board is in my brewing temperature control machine, and it would be inconvenient to lay a cable to it, let alone arrange for the board to be shared between the temperature control software (2 relays) and the sprinklers (up to the remaining 6), so I'll probably end up installing another machine with a new board. Conveniently, the machine needs to be located in the same place where my old access point used to be located:
|
|||||||||||||||||||||||
This photo also shows the open lid of the dead sprinkler controller at bottom right. The network machine is now long gone, so there's a spare network cable there.
| Tuesday, 28 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Spent some time in the morning trying to set up a diskless machine sprinkler.lemis.com for running the sprinklers. Writing the program was simplicity itself: a couple of functions from the brewing temperature control software for accessing the relay board, and a main loop that just sets the relays and sleeps. The rest of the job is done by cron, of course.
Unfortunately, setting up the machine wasn't as easy—the same old story. The machine I had chosen had a defective disk controller, so I had two reasons to want it to be diskless (the other is that disks are relatively noisy and heavy on power consumption). In addition, I didn't have a spare IDE disk for the installation, so tried with an old SCSI drive, only to discover that most of my old host adaptors didn't want to supply the boot disk. Finally got it installed and the old dual processor Pentium II machine that I installed with no longer wanted to recognize the disk. Put the host adaptor in the target machine and it reliably paniced on boot; maybe there's more wrong with it than the disk controller. Never mind: old machines of this calibre are a dime a dozen, literally.
While playing around with this, Dan Carosone and his wife showed up for a lightning visit, shortly to be followed by Bernd Wulf, so spent some time with them. After they left, I still had no time: had to cook the turkey that we bought for Christmas.
For some reason I've been having a lot of trouble with meat thermometers. The ones available in Australia are calibrated in Fahrenheit, with an afterthought Celsius conversion of the 10° F steps thrown in. That drives me mad, so in Germany last month we bought one with proper calibration. It disintegrated after the first time we used it.
Last Friday I found something similar in town, but it was designed for heating milk. Then I found the ultimate solution: a digital meat thermometer (also timer and clock) with a long sensor, capable of measuring temperatures up to 260°. Today I confirmed that it is useless: the metal shielding conducts enough heat to the sensor as to completely falsify the readings. sigh.
Still, the turkey came out OK (with the help of the old, almost illegible thermometer that I've been trying to replace all this time), and we all ate too much.
| Wednesday, 29 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Back to work today, along with attacking the long mail queue that had built up over the Christmas weekend. Did some performance testing (on my laptop no less!) which appeared to be nearly double the speed of the tests we had been doing at work. We're obviously missing something here.
| Thursday, 30 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Another frustrating day. Set off to look at the profiling stuff again and found that the Makefile (called makefile, though it should have been called GNUmakefile) was building profiled objects ending in .o and linking them with non-profiled libraries. Under these circumstances, it's not surprising that we're getting funny results. Spent some time looking at that, but even after changing the object names to end in .po, it still built .o files, and then complained about the lack of .po for the linker phase. These appear to be default GNU make rules; I need to find out more about what has happened to GNU make since I last used it in earnest over 10 years ago.
It also became clear that I need to run Linux on one of my test boxes. I had borrowed some CDs of Fedora Core 3 from Tim Stoakes last week, and tried to install them on duet. First, though, I had to remove the old image, and since that was a good candidate for the sprinkler system, set up a network boot version. The result, with the target machine, was a halt in the BTX loader. Tried it on another machine, where the kernel load works perfectly. I'm going to have to admit that the motherboard has had it: the onboard IDE controller can't access IDE disks, it panics with a SCSI host adaptor, and it doesn't even get that far with a network boot.
Installing Fedora 3 proved to be only marginally easier than installing Fedora Core 2. It found fault with Tim's CD-Rs, though it continued and died at some later point without any specified reason. I read in the CDs on eucla (with no problems) and burnt new ones. A reinstall of the existing system was a complete catastrophe: it appears that, since no kernel was on the disk, it didn't bother to install one. I had to start again from scratch, and note once again that when I told “Disk Druid” to use all the remaining space for swap, it used exactly half of the remaining space (about 300 MB). At any rate the installation finished, and it didn't have any trouble mounting the CDs I burnt on FreeBSD, so I suppose things are looking up.
In the evening considered alternatives to my broken 120 MHz Pentium motherboard for running the sprinklers, and found an old Sun SPARCStation 1, which seems to have everything necessary. It's a sign of the times, I suppose, that what used to be a high-end workstation 15 years ago has been degraded to a sprinkler controller, but I suppose it'll be more reliable than the Pentium.
| Friday, 31 December 2004 | Echunga | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
Another day where I seem to have got nothing done. Gradually I'm becoming very frustrated by the way most people write mail. Never mind the controversy between “top posting” and “bottom posting”: both are wrong, and it shows. Spent a lot of time tidying up people's mail messages today, and in almost every case saw confusion that arose from the way they wrote their messages. We have the tools: why don't people use them? In particular, leaving irrelevant text lying around, coupled with putting answers a long way from the (unidentified) text to which you reply, is a recipe for disaster. Time to update my mail usage guidelines.
Apart from that, sort of got the Fedora Core 3 installation finished. It started up the final setup in graphic mode, just what I want for a storage systems test machine, and then rebooted. And the display didn't work. Spent quite some time confirming that the X configuration (which really used bog-standard parameters) was correct, and that it wasn't a problem with the old monitor I was using. It really looks like X is confused by the display card (an old Cirrus Logic). Gave up and ran the thing in text-only mode (which makes more sense).
Why is it so difficult to install additional software on Linux? Fedora has a thing called anaconda which will give you a (too brief) overview over what to install—before the system is running. After that it no longer seems to be available, and the alternative (yum) is not exactly a replacement. And, of course, most of the tools that come in the base FreeBSD system are missing. sigh. Just getting a Linux box to a usable condition seems to take forever (though admittedly I could have saved a lot of time if it had been able to reliably read the CD-Rs).
Do you have a comment about something I have written? This is a diary, not a “blog”, and there is deliberately no provision for directly adding comments. It's also not a vehicle for third-party content. But I welcome feedback and try to reply to all messages I receive. See the diary overview for more details. If you do send me a message relating to something I have written, please indicate whether you'd prefer me not to mention your name. Otherwise I'll assume that it's OK to do so.
| Top of page | Previous month | Greg's home page | Today's diary entry | Next month | Greg's photos | Copyright information |
