|
| Greg's diary |
| July 2010 |
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
This view of the diary is limited to these topics: technology. There may be lack of continuity in the text, and some days may be completely missing. In case of doubt, please enable the complete display.
| Thursday, 1 July 2010 | Dereel | Images for 1 July 2010 |
| Top of page | ||
| next day | ||
| last day |
|
FreeBSD file deletion without soft updates
|
Topic: technology | Link here |
Last month I noted the extremely long time it takes Linux to delete large files. It took nearly 30 seconds elapsed time to delete a 190 GB file, while FreeBSD does it in a few milliseconds. One hypothesis was that FreeBSD's speed was only due to soft updates, and there's some reason to believe that. Today was the first of the month and thus the time for my monthly level 0 dump of my file systems. I back up to disk, so I first needed to make space by deleting a number of large files, 36 GB in total. They're on a separate file system, so I was able to umount it, reset soft updates, mount it again and delete the files. I then set soft updates again, copied a similar backup and deleted it. Here the results:
No soft updates:
Soft updates:
So yes, it's much faster with soft updates than without. Interestingly, the processor time also drops to almost 0; I suspect most of the time is in non-process context, which time(1) can't measure. But the time without soft updates is about 0.062s per GB. For Linux (XFS) it's 0.63s per GB, 10 times as much. My (possibly inaccurate) recollection is that it was even slower with ext3. There's still something puzzling about the figures.
There are differences in the scenarios: I think my FreeBSD box is faster than the Linux box, though not much, and I was deleting a number of smaller files. But I don't think that that is particularly relevant. It's clear that the CPU time is not an issue in the equation (indeed, the slower Linux box used only 1.112s CPU time to delete 190 GB, while my faster FreeBSD box needed 0.800s to delete the 36 GB). The disks are effectively the same speed. I can't see any reason why it should be faster to delete a number of files than a single file of the same size; I'd expect it to be slower due to other overhead.
|
One-step HDR panoramas with Hugin?
|
Topic: photography, technology | Link here |
Hugin has a number of poorly documented features, one of which is the plethora of choices on the Stitcher page:
|
|
||||||||
It occurred to me that it would be nice to merge the “before” and “after” photos of the gum tree removal to get an exact view of the change. That should be possible with programs like align_image_stack, but my previous attempts with it proved to be less than successful. Some of the choices on the Stitcher page looked like they would do what I want.
If they do, I couldn't find it. But some of the other choices looked interesting, particularly “Merged and blended panorama”. Currently I do this in two steps: first I create HDR images of each view, and then I stitch the HDR images. It would be nice to do it automatically. Tried that with one of last Saturday's panoramas, without success:
Maybe I can drag out some more information about the topic.
| Friday, 2 July 2010 | Dereel | Images for 2 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Which data projector?
|
Topic: multimedia, technology | Link here |
Over the last few days I have gradually come to the conclusion: I need a 720p projector (1280x720), not a 1080p (1920x1080), since that's what the TV stations broadcast. Today spent nearly the whole day looking for a suitable projector. What I found was:
As I had expected when I bought the last projector, the standard projector now is 1080p.
Those 720p projectors still on the market are either old (such as the Panasonic PT-AX200E) or not really intended for home theatre use (such as the Epson TW450, which has a real resolution of 1280x800 and significantly higher noise levels).
A limiting factor for my particular application is the size of the room: it's not very wide, and there would only be about 3 m between the projector and the screen. Only some projectors can project an adequately sized image at that distance.
But which ones? I know my old Panasonic PT-AE700E can, and I downloaded the manual for the PT-AX200E, which confirmed that it had exactly the same parameters (probably the same lens as my 5 year old PT-AE700E). But what of the others? The documentation on the web sites has hit a new low. For example, the “spec sheet” for the Epson TW450 states:
| SCREEN SIZE (PROJECTED DISTANCE) | 33 to 318 inch [0.9 to 9.0m] (Zoom: Tele), |
| 33 to 317 inch [1.08 to 10.5m] (Zoom: Wide) |
What on earth does that mean? Our best guess on IRC is that the inches measurements are the screen diagonal and the metric units are the distance, which would mean that you can get a 33" display at distances between 90 cm and 108 cm (convenient metric values that presumably just happen to correspond to 3' and 3' 6"). What kind of mind can create that? A mix of units, inadequate description of the meaning, and a representation that requires significant calculations to work out what the size would be at a given distance. If I take that trouble, I discover a maximum diagonal of 2.8 m at 3 m distance. The Panasonics have 2.5 m, so that would seem OK. But is my guess correct?
This would be terrible enough if only Epson did it, but I've seen other manufacturers do it as well. Sure, you don't need a university degree to become a web master (I suspect it might even be a disadvantage), but this complete mess makes me wonder if mind-altering substances are a requirement. Wouldn't it be easier to say that at distance n the diagonal can be between 0.79 n and 0.92 n?
Projector bulb life is very variable. The Epsons offer up to 4000 hours, while the Panasonic PT-AX200E has a hard limit of 2000 hours, after which it refuses to work (OK, you can reset the life timer, but they warn about exploding bulbs, just as they did for my old projector).
Warranties are limited. Panasonic only has 1 year (particularly significant in view of the damage that happened to my old projector after about 18 months), and there are some interesting clauses: not more than 4 hours continuous use. This relates to the optical block, which is where my projector was damaged, so I assume the Panasonics have a weakness in this area. Epson has 2 years' warranty, and Sanyo has 3 (which according to Ausmedia also includes the globe), though their web site doesn't mention anything about the warranty terms.
By evening I still hadn't got much further. Did some experimentation and decided that 1080p projectors don't seem to have difficulties displaying 720p, so it might make more sense to bite the bullet and get an up-to-date projector than buy something that is no longer even state of the art.
Yet another issue with Panasonic became apparent in the evening. The X display didn't work; instead of 1280x720, I only got 640x480. Further investigation showed that the projector doesn't supply EDID information, and without mode lines X assumes that it can't use higher sync frequencies than VGA. This problem didn't hit us yesterday because the Acer projector was connected when I started X. Worked around that by reconnecting the Acer projector, starting X and then swapping projectors. Here the relevant diffs from /var/log/Xorg.0.log:
In addition, I was reminded the hard way that the projector can't synchronize to the frequencies that FreeBSD uses for the console, which is slightly higher than normal VGA. I doubt that they have invested much in the newer 720p modes. There are clearly enough reasons not to buy a PT-AX200E: limited warranty, poor compatibility, previous reliability issues, low lamp life and relatively high price. The cheapest price I found for one is $1970; for $2170 I can buy a Sanyo PLV-Z700 1080p unit with three years' warranty and a short-focus lens slightly better even than the Panasonic. About the only down side (apart from the fact that I didn't really want a 1080p) is that the lamp is relatively dim at 1200 lumens. But that's still brighter than the Panasonic I have now (1000 lumens).
|
Linux.conf.au in Ballarat
|
Topic: technology | Link here |
The Ballarat Linux Users Group is interested in hosting linux.conf.au in 2012. That's quite a challenge: so far it has only been hosted in big cities. The smallest in Australia was Hobart, with a population of 220,000, and that proved a problem at the time. Ballarat has less than 80,000 inhabitants. That reflects itself in the number of people available to organize the conference, and also in the range of venues which we could use. This evening Josh Stewart organized a first meeting at the Thai Sala Pavilion Restaurant in Sturt Street. As recently at my weather station presentation, the size of the town was evident: we had only 5 people, one of whom still needs to develop the sense of responsibility needed for such an event:
|
|||||||||||||||||||||||
Still, I think we should be able to manage it.
| Saturday, 3 July 2010 | Dereel | Images for 3 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Digital devices: data moves the world
|
Topic: technology, opinion | Link here |
We're still playing around with the navigator, which from time to time asks us to go ways that we wouldn't normally take. Today we tried some of them; it's not clear whether they're better, but it's probably worth trying it out.
One of the problems is data accuracy. On the way into town, at one point we were told to turn second left. Only one street. Drove past and were told “Recalculating”, in other words, “you blew it”.
On the way home we stopped off in Napoleons at the little roadside plant sales place. Didn't buy anything, and since we were about 50 m from the main road, we should have turned around and continued on that road. But the navigator wanted us to continue over a kilometre and take a right turn into this road:
|
|||||||||||||||||||||||
It's not clear why: it's clearly longer than turning back. Possibly it had other tricks up its sleeve. Clearly a navigator can only be as good as the map data, and this is deficient. But it reminded me of a recent Microsoft advertisement in issue 10/2010 of c't, with a claim:
In English, that's “Code moves the world”, and at the time I marvelled at the stupidity, though it's in line with my experience with Microsoft MUAs. Looking for it now, it's no longer there, not even on the web site; Google draws a blank. So maybe they have realised how stupid the idea is and removed it again. The only hit I got in English was more sensible:
... the idea is that content and not just code moves the world
Clearly this experience helps relativize that claim.
| Sunday, 4 July 2010 | Dereel | Images for 4 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
UNIX history: RFC for networking
|
Topic: technology | Link here |
While looking for something completely different today, came across RFC 681, dated 18 March 1975. It appears to be the earliest use of UNIX (Fifth Edition) in an (ARPA) network environment. The most interesting thing from my point of view is that they use the standard file system interface to talk to the network. It's a pity it doesn't give more detail.
|
HDTV: Worth the trouble?
|
Topic: multimedia, technology | Link here |
The other day I did some comparisons of the quality of a 720p broadcast stream displayed at natural resolution and also scaled up to 1080p. We discussed it on IRC today, and I took some screen shots. The result? What I sort of knew already: the quality of Australian HDTV isn't really very good. We're using MPEG-2, which makes relatively inefficient use of bandwidth. What's the bandwidth? According to the transcoders, it's 90 Mb/s, or 11.25 MB/s. But that would give a recording size of 40 GB per hour. In fact, a typical 35 minute recording uses 2358107876 bytes, or 1.123 MB/s—a number suspiciously close to 10% of the nominal rate. The results are clear: digitalization artefacts, particularly in scenes with motion in them. The following two images are 720p and 1080p, and they really need to be looked at full size:
|
||||||||||||||||||||||
|
||||||||||||||||||||||
There are many artefacts, but the ones around the scoreboard are particularly obvious. Clearly any loss in the scaling pales by comparison. I hear from friends overseas using MPEG-4 and H.264 that the results are much better.
On a different topic, updated my X configuration to work with the Panasonic projector. That was easy: I had commented out the entry, so I just needed to change it. But the image quality got much worse. Further examination showed that the mode line I had chosen generated a different horizontal frequency. Here the “old” (Acer) and the “new” (Panasonic):
The first generates a horizontal frequency of 59.8 kHz, and the latter one of 45.9 kHz. Both are in spec, but it seems that the Panasonic doesn't handle them equally well. After the change, the image quality was much improved.
| Monday, 5 July 2010 | Dereel | |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Wikipedia down!
|
Topic: technology | Link here |
Round here there's nothing unusual about getting an ECONNRESET error loading web pages, and when I got one with Wikipedia today I just assumed that it was the normal flakiness of my satellite link. But this time it stayed, and it proved to be a power failure in their primary data centre. The interesting thing is that the details are all available online at URLs like http://ganglia.wikimedia.org/ and http://nagios.wikimedia.org/, though the latter seems a little behind the times. Spent some time watching that until we realised that the system was back up, but Nagios still claimed lots of down systems.
| Wednesday, 7 July 2010 | Dereel | Images for 7 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Setting up the new projector
|
Topic: multimedia, technology | Link here |
Spent some time setting up the new Sanyo PLV-Z700 projector today. It's much bigger than the old ones. Here the Acer, the Panasonic and the Sanyo from top to bottom. The backs of each projector line up.
|
|||||||||||||||||||||||
Things started off well: plug in, turn on, and it displayed a perfect image—in 1280x720, because that's what my X configuration said. Getting it to work correctly in 1920x1080 was a very different matter. Decades ago I wrote the book when it comes to setting up X mode lines, but in those days displays were invariably CRTs. Times have changed, and projectors seem to have a lot of idiosyncrasies that even LCD panels don't have. Today X ignored all my mode lines with the message:
Why? I had a mode line:
Normally X tells you why the mode lines aren't valid (“Horizontal sync out of range” or some such) but here there was no explanation. Checked all the parameters, all the messages, and made no progress. Finally ran X -configure and took the generated xorg.conf file. That worked, but the picture was positioned far to much to the right. xvidtune didn't help much; I gradually moved the image to the left, but it got dimmer in the process, something that looked worryingly like the way to kill a monitor in the bad old days.
The handbook (conveniently not included in the 220 page book included with the projector; that's a “Quick Reference Guide” and is 20 pages × 11 languages) tells me that it will basically accept any signal up to 100 MHz dot clock, but it gives as the 1080p frequencies 67.5 kHz/60 Hz (for USA and other compatible countries) and 56.25 kHz/50 Hz (for the rest of the world). The absolute theoretical minimum dot clock for 1920 dots at 67.5 kHz is 130 MHz, so it's not clear what that statement can mean. Ignored the dot clock limitation and tried to set the horizontal and vertical frequencies, not without some surprises:
|
|||||||||||||||||||||||
The information display on the left doesn't seem to even come close to the values that xvidtune and I expected: this image was at 56.31 kHz/50.28 Hz, but the projector claims it's 52.1 kHz/46.6 Hz. And to get that I have a pixel clock of 132 MHz, way outside the claimed range. Played around for a while and got an acceptable image, but the image quality looks like it could do with improvement.
Spent some time looking for alternative sources of mode lines. The first one started with a reference to this diary, as did a one more specific to the projector. Gave up and considered the Microsoft space. How do you convert data in a Microsoft “driver” to a mode line? The simplest way seems to be to read the data out of the GPU registers, if that's possible. Investigated a little, and on a suggestion looked at the CD that came with the projector. It contained only the manual—no software at all. That makes sense—I never understood why Microsoft space devices always require a “driver”—but I think that's the first time I've seen a display device that comes without one.
On recollection, it's not that surprising. Projectors and digital TV displays take standardized signals, for example from DVD players or TV broadcast, so the timing must be standardized. All I need to do is to find the standard. In the meantime, found a reference to a 1080p mode line for a different display in the MythTV modeline database.
It looked wrong: the sync pulse was too close to the end of the display, which should have pulled the display to the right, but when I tried it, it worked relatively well. So I now have the thing running, and can spend more time tuning to perfection.
And what's it like? The lens doesn't seem as good as the Panasonic: there's some chromatic aberration, which is visible in the photo above. But the colours are an order of magnitude better than the Acer, and the brightness is good, far more than I expected for a projector with only 1200 lumens. Some time I must measure the current brightness of all three projectors.
|
No resolution from SkyMesh
|
Topic: technology | Link here |
SkyMesh is taking forever to resolve my satellite link problems. Each message to support takes two weeks to get a response to, and I'm no longer prepared to accept the claims that the problem is a defect in the modem. They pointed to the “power on” events as “evidence” for that, but they haven't been able to prove it. Things are not getting any better, so sent off a message to Paul Rees, the manager, who in the past has been helpful, and suggested that if he thinks it's the problem, they should install a new modem and prove it. Got a reply back (Microsoft-style, appending the message rather than answering it, and thus missing several points) telling me that they didn't want to spend the money, and that I could get a modem elsewhere if I want. He also appended the log of the most recent outages, only one of which was a “power on” (and that was me power cycling it). It looks as if I'm going to have to complain to the DBCDE or the TIO.
| Friday, 9 July 2010 | Dereel | Images for 9 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
| Saturday, 10 July 2010 | Dereel | Images for 10 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
New USB tuner: success
|
Topic: multimedia, technology | Link here |
Finally got round to putting the new (Winfast DTV Dongle Gold (who thinks up these names?) into cvr2. It worked out of the box—that's a first—but for some reason I couldn't persuade MythTV to do a manual recording on it, possibly because of some incorrect timing calculations. Shortly afterwards it recorded a film on it normally, so there's nothing wrong with the tuner.
About the only issue I had wasn't directly related to the dongle: it only has an antenna in connection, not an antenna out. One of the two existing (PCI) tuners I have in the machine also has no antenna out, so there was no direct way of connecting all three. Over to Chris' place to collect the third one, but it proved that she didn't have it, and I spent a fair amount of time looking for tuners and also a display card with TV out to connect up to the analogue TV I lent to her a while back. This is the same configuration that I set up on 18 September 2004, and which I described in some detail. At the time I wondered whether it wasn't too much detail, but now I'm glad I did it.
| Monday, 12 July 2010 | Dereel | Images for 12 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
SkyMesh connection problems
|
Topic: technology | Link here |
My discussion with SkyMesh is going nowhere. My support requests go unanswered for up to 2 weeks. They claim that my modem is defective because they can't reach it from time to time and ignore the fact that these issues have occurred with three different modems. At the same time, they want me to pay up front for a very expensive replacement. When I asked why, I got the response (original formatting):
This is tantamount to accusing me of intending to steal the modem. The facts of the matter look very different to me. Since moving to SkyMesh I have had 125 outages, measured as when both the other end of the link and all 5 systems that I ping are unreachable. I add this latter check because it seems that sometimes the other end of the link doesn't respond. But I've been seeing a lot of this kind of thing:
The interesting columns here are the second (ability to ping the other end of the link) and five systems spread around the world. In this example, the link appears to be down, but I'm able to ping random subsets of the five systems. This looks very much like problems in SkyMesh's network to me. Wrote back and asked them to fix that first, and then we could see what was wrong with the modem. I'm losing hope of an amicable settlement.
|
w_scan revisited
|
Topic: multimedia, technology | Link here |
w_scan is a utility to perform a frequency scan on DVB-T tuners. I've tried it a couple of times in the past, but it didn't work: it scanned only the frequencies stored in a pre-defined table, and they weren't the Australian ones.
That seems to have changed: it now takes a parameter to specify the country, so I tried it out again. The result: it works some of the time. It doesn't find anything at all with the new Winfast DTV Dongle Gold, and with one of the other tuners it didn't find SBS:
The problem here is that SBS broadcasts on 634.625 MHz and not 634.500. w_scan sees the real frequency, but drops it. The filter timeout seems to be part of the issue.
| Thursday, 15 July 2010 | Dereel | Images for 15 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
ALDI Medion ΛΚΟΥΛ netbook
|
Topic: technology | Link here |
This week ALDI had a netbook on special, only $389, with a real keyboard, 250 GB disk and 1 GB memory, and the incongruous name Medion Λκουλ (which friends claim is really something written with fonts that imitate Greek). ALDI hardware has the wonderful advantage that if you don't like it, you can return it for a refund within 2 months, clearly a great advantage considering the hardware compatibility issues that free operating systems have with small laptops.
Yvonne brought back one, and I spent the afternoon playing around with it. It doesn't have a DVD drive, though it comes with a number of what I think are DVDs, one of which has the interesting statement (in German) “You are not authorized to make unauthorized copies of this data medium”.
So I need a USB DVD drive. Fortunately I bought a new drive last week, and I have an old USB adaptor housing. Put the drive in and tried things out; worked. But it wasn't the disk I wanted, so I still had to burn one. Put that in swamp, my test machine, and had lots of difficulty, finally coming to the conclusion that Yet Another Motherboard has a defective USB. I couldn't mount my external USB disk drive there (it stopped the machine from even finishing POST), and the USB keyboard had trouble when anything else was connected. Damn. That's the last of that series of motherboards, so now I have plenty of perfectly good left-over memory and processors. I wonder if I can find old motherboards on eBay at a reasonable price.
The next attempt was with a USB stick, which I burnt in teevee. It wouldn't boot; it looks as if the image was faulty. But I did locate a DVD with FreeBSD 8.0-RELEASE, and booted from that. The first thing I wanted to do was to back up the disk, which had to go over the network, so went into the Fixit mode. The kernel found the (wired) Ethernet interface, but not the wireless one. And after a lot of investigation, it proved that I couldn't receive any data via the interface. netstat showed that the data was arriving, but it didn't seem to make it out of the driver.
In the end put the disk from swamp (9-CURRENT from some time in March) into the USB housing and booted the netbook from that. That worked, and so did the wired interface; still no wireless interface, but I'll look at that problem later. Copied the disk image to dereel, compressing it from 232 (“250”) GB to 17 GB in the process, and showing that these Atom CPUs really aren't very fast. Tomorrow I'll install the image on the internal disk and investigate getting the wireless interface going.
| Friday, 16 July 2010 | Dereel | Images for 16 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Building world on Λκουλ
|
Topic: technology | Link here |
As planned, copied the external disk to the internal disk on the Λκουλ netbook. In the process, discovered a complete object tree for the last OS build, clearly nothing I needed to copy across, so during the copying (with bsdtar), removed the remaining parts of the /usr/obj tree. Surprise: unlike gnutar, bsdtar can't handle that. It died on me without completing the remainder of the copy. That looks like a serious bug to me: sure, people don't normally remove files while they're being copied, but it can happen. Another reason to stick to gnutar (the other is that I'm more used to the baroque command line options).
After that, as planned, brought the system (FreeBSD CURRENT of about March of this year) up to date. That took forever:
That's over 4½ hours elapsed time, nearly 7 hours user time and 83 minutes system time: over 8 hours CPU time in total. It would be nice to blame that on the WITNESS option that is set by default in -CURRENT, but that should only affect the system time. The times for the kernel build were corresponding, and even the installworld target (mainly copying) seems to have taken longer than usual, possibly because of the low-power disk:
This is all without a valid comparison, of course, but next time I build on a “normal” computer, I'll be able to do that comparison. More to the point, I still don't have a wireless interface, and I can't start X:
That's not surprising, considering the age of the X server, but I think I'll try Linux first before I upgrade. If I can't get the wireless interface working, it won't help anyway.
|
Reinstating the mystery plants
|
Topic: gardening, technology | Link here |
What is that flower that I photographed yesterday? It was one of my mystery plants, but then I found out what it was and moved it from that page. Spent a while searching for it (as I mentioned in yesterday's entry, it's chasmanthe floribunda), but it occurred to me that I hadn't done anything sensible by removing the entries from the mystery plants page, so spent some time reinstating them. One of the issues was that I had changed the names of the pages (for example, http://www.lemis.com/grog/Gardening/mystery-12.php became http://www.lemis.com/grog/Gardening/chasmanthe-floribunda.php), and it's possible that some references remain to the “mystery” name. Sure, I can put in redirects, but how about a single file that contains the lot? In the end I wrote a page that looks like this:
The function redirectme looks for the name of the page in the first column and redirects to the page in the second column. All I needed to do was create a link to this page with the name in the first column. No extra files (though I did use a symlink so that I can update the page easily).
|
IPStar modems: disconnect where possible
|
Topic: technology, opinion | Link here |
Sundance promptly fired up his computer and tried to get a DHCP lease. Problem: we don't use DHCP. But the satellite modem has a server, so enabled it. Result: reboot and a minute off the network—a programmed dropout:
That took 48 seconds, with an uncertainty of about 30 seconds—noticeably less than the other dropouts. Does this mean that the others have some other reason?
| Saturday, 17 July 2010 | Dereel | Images for 17 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Ubuntu on Λκουλ
|
Topic: technology | Link here |
As planned, installed Ubuntu 10.4 on Λκουλ. Why do Linux installations take so long? Anyway, it came up and recognized the display hardware immediately, and the Ethernet interface was also recognized. But still no joy with the wireless LAN card. Did some googling and discovered that yes, people had had it working, but I didn't have enough time to follow through.
About the only thing that did become apparent: this machine has a 1024×600 display, which I knew before I bought it, of course. And they seem to be becoming more prevalent. All the more reason to reject window managers and programs that take up the top and bottom of the display. After removing all the bars and status displays and things, the area remaining for a “full-screen” firefox window is only 1012×376, barely more than half the height of the screen, and barely more than the display height of the EGA that IBM introduced nearly 30 years ago:
|
||||||||||||||||||||||
|
||||||||||||||||||||||
So modern display software wastes 38% of the usable area on this tiny display. That's yet another reason to reject GNOME and other Microsoft-like display environments and use a window manager that doesn't want to be an entire computer.
| Sunday, 18 July 2010 | Dereel | Images for 18 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
More fun with Λκουλ
|
Topic: technology | Link here |
So I have Ubuntu 10.4 running on Λκουλ, but without the all-important WLAN interface. Upgraded the kernel to the latest version, without any obvious improvement, and then spent more time looking around the web and found some instructions on how to compile the driver. Did that—another 200 odd MB of data—and once again there was no improvement.
OK, the reason I tried Ubuntu is that it might have been easy. I always find so many things that grate on Linux, and for some reason that I didn't care to follow, the NFS file systems would no longer load—instead I got the startling message Disconnected from Plymouth, which doesn't seem to have any further meaning. But the text-mode terminals were missing. I'm not sure why, but using this tiny interface with the funny mouse to do anything is so painful that I couldn't be bothered. Back to investigate FreeBSD.
But how? Ubuntu had not been able to install its bootstrap in the root partition, so it had overwritten my FreeBSD boot selector. OK, I've used GRUB before, so went off looking for the config file. Surprise, surprise! The authors had recognized that GRUB was a pain to use, so they rewrote it and replaced it by something that is an order of magnitude more painful to use. All I wanted was to add an entry for the FreeBSD partition on the disk, which should have been a couple of lines. To ensure a smooth transition, they've changed the names of the configuration file, but finally found the official manual for GRUB 2, conveniently numbered 1.98-r2508, and discovered that the configuration file is now called /boot/grub/grub.cfg, which starts of like this (somewhat trimmed):
So it's no longer a configuration file: it's a script. At the very least they could have separated configuration information and code. Took a look in /etc/grub.d, which, if anything, was worse. It's all just scripts.
Finally found instructions, to do exactly what I wanted: add FreeBSD to a Debian boot. What I ended up with was a file /etc/grub.d/40_custom with the following content:
The meaning of chainloader is not really clear from the manual, which doesn't give any BSD examples, though it does state that chainloader is not necessary for them. I also didn't remove the comment, though I should have done, since it's completely wrong. This isn't easy. The idea of modular configuration files is arguably useful, but mixing code and data in a “configuration file” is most certainly not. To quote one of the comments in the previous instructions, written by “doik”:
i install grub 2. it is foul beyond words to configure and is not pretty at all. it also seems much slower at getting to the menu. grub 2 is FAIL.
So, finally I had a GRUB configuration that should boot FreeBSD. Rebooted. No menu popped up, and I ended up back in Linux. Read the manual again. Played around with variables that might potentially change the behaviour. No change. Finally, Juha Kupiainen told me that I had to hold down the left shift key when booting. That worked: I got a menu and was able to select FreeBSD. But there's nothing anywhere in the manual that mentions this. Somebody told me that this is a (presumably undocumented) Ubuntu extension. What a pain!
Still no wireless card, of course. pciconf -lv tells me:
I wonder what the question in brackets is intended to mean. More googling, and found instructions for installing this code, first requiring downloading the drivers from the Realtek web site, which led me around a couple of cycles in a link loop before starting to download two copies. Left that until tomorrow.
| Tuesday, 20 July 2010 | Dereel | Images for 20 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
GRUB 2 revisited
|
Topic: technology | Link here |
I haven't been playing with λκουλ much lately—I'll take a stab at the NDIS-based wireless network later—but I wanted to boot it in FreeBSD today to try out something potentially dangerous. And, of course, the Left shift trick didn't work: I had no way of booting into anything but Linux.
This is with GRUB 2, and there's supposed to be a way of getting it to present the boot menu automatically, like there was with GRUB 1. But everything has changed. The files are spread across three directories: global information is in /etc/default/grub, “configuration” information (consisting largely of scripts) in /etc/grub/, and the boot environment is in /boot/grub/, all 186 files and 4.3 MB of it. I've understood some of the puzzle, but not how to get the menu. It looked as if it should be an entry in /etc/default/grub, but I couldn't see one. There are comments for some of the values, like the all-important resolution of the graphical display—for a boot loader!—but not for the menu. This is clearly a case of FTFM: it's almost impossible to find out how to configure the thing.
With the help of a number of people on IRC, finally got it to work. Ubuntu has its own document, apparently written as an afterthought, and at the section with the emetic URL https://help.ubuntu.com/community/Grub2#/etc/default/grub%20(file)—with spaces in it!—I found the answer: undefine GRUB_HIDDEN_TIMEOUT. It seems that this should have happened automatically: one of the myriad scripts, /etc/grub.d/30_os-prober, should have detected FreeBSD, but it didn't. So, finally, after a couple of hours, I can dual boot the machine the way the FreeBSD boot0 boot manager would do without any configuration. Isn't progress marvellous?
|
More network issues
|
Topic: technology | Link here |
Access to http://www.lemis.com/ failed today. It's hosted at TransACT, and further investigation showed that I couldn't access that network from 09:54:17 until 10:33:16, nearly 40 minutes. Traceroutes from this end ended at Equinix:
=== grog@dereel (/dev/ttypa) ~/public_html 225 -> traceroute ozlabs.org
I was able to access TransACT from my US server, so it wasn't TransACT itself. A traceroute in the other direction looked like:
=== grog@bilbo (/dev/pts/2) ~ 2 -> date; traceroute sat-gw-ext.lemis.com
Probably not SkyMesh's fault, but they're the only people I could report it to. I'm too angry with them at the moment to call them up and risk losing my temper, so sent a mail message at 10:23:51. No response, no acknowledgement of receipt. I wonder to what extent the resolution of the problem was related. Certainly that would mean that they reacted pretty quickly (and that thus the problem was at SkyMesh after all).
|
Powercor: killing uptime
|
Topic: general, technology | Link here |
dereel has been up for 110 days now. Or it was. While watching TV in the evening, we had a power failure, this time over 3½ hours. Yes, we now have UPS running just about everything, but they don't last that long. At least we didn't damage the new projector.
| Wednesday, 21 July 2010 | Dereel | Images for 21 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Λκουλ: installing NDIS
|
Topic: technology | Link here |
Continued trying to get the wireless network card running on λκουλ today. I had downloaded the driver from the Realtek site, and the instructions were straightforward enough. But where were the drivers? The instructions mention the files net8192se.inf and rtl8192se.sys, but what I got was a whole directory tree. Went finding and found four files, RTL8191_8192_SE_WindowsDriver_2016.2.0521.2010.F0062_23.P0525_ISS_1.00.0157.Win7.L/91_92_SE_Driver/Win7X64/net8192se.inf, RTL8191_8192_SE_WindowsDriver_2016.2.0521.2010.F0062_23.P0525_ISS_1.00.0157.Win7.L/91_92_SE_Driver/Win7X86/net8192se.inf, RTL8191_8192_SE_WindowsDriver_2016.2.0521.2010.F0062_23.P0525_ISS_1.00.0157.Win7.L/91_92_SE_Driver/Win7X64/rtl8192se.sys and RTL8191_8192_SE_WindowsDriver_2016.2.0521.2010.F0062_23.P0525_ISS_1.00.0157.Win7.L/91_92_SE_Driver/Win7X86/rtl8192se.sys. Presumably the Win7X64 files were the ones for 64 bit mode and the Win7X86 are for 32 bit mode (and not 86 bit mode, as the convention would seem to imply). But what about Win7? That looks like Microsoft “Windows 7”, which is not what I need (though it's what came with the machine). Tried running ndisgen against them, and it happily generated a KLD that I could load. With verbose boot messages, it also produced quite a bit of information:
But that was all. It should have generated probe messages for an interface ndis0, but nothing happened. So what's wrong now? Wrong drivers? Or something else? It would have made more sense to take the drivers that came with the machine, but they're happily packed up in a disk image. It seems that I could use mdconfig on it if it weren't compressed, or geom_uzip for it if it had been compressed with mkuzip. The image is 250 GB in size, so it would be nice to be able to pipe the output of bzcat into mkuzip, but mkuzip wants a disk file or a character device. It was easy enough to hack it to read from stdin, but it wants to issue a call to stat to find the size of the file. A little more work and another flag should solve that problem. Manãna, at the earliest.
|
Australian Broadband Guarantee: help from the DBCDE
|
Topic: technology | Link here |
There doesn't seem to be an obvious solution to problems I'm having with SkyMesh, so today I called up the Department of “Broadband”, Communications and the Digital Economy to ask for help. The number they gave was 1800 254 649, but though that's the Department, it's the wrong number. The correct number is 1800 883 488. Called that and got a recorded message telling me that they were too busy to take my call, that I should leave my phone number and they would call me back. I did, they didn't.
| Friday, 23 July 2010 | Dereel | Images for 23 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Λκουλ: progress towards WLAN
|
Topic: technology | Link here |
More playing around with the WLAN issues on λκουλ today. Spent some time discussing where the firmware was; it's not in the distribution I got from Realtek, and the instructions that I had didn't mention any, but the documentation does mentions firmware. Callum Gibson found some somewhere, but it wasn't clear if it matched or not. He also found a Microsoft “Windows XP” driver for the chip set, and Sue Blake sent me the drivers she had on her identical Λκουλ netbook. Tried Callum's first, without success—until I realised that kldload rtl8192se_sys does not respect the current working directory; instead it uses the value of the sysctl kern.module_path unless a directory component is specified.
Finally got the correct module loaded, and—success!
And I had an interface:
=== root@swamp (/dev/pts/0) /src/FreeBSD/NDIS/Realtek-8191/realtek-XP/guts 110 -> ifconfig ndis0
No carrier? Tried bringing it up, but it seems that that's not the way you do it any more. Instead you need to create the real interface:
=== root@swamp (/dev/pts/0) /src/FreeBSD/NDIS/Realtek-8191/realtek-XP/guts 113 -> ifconfig wlan0 create wlandev ndis0
=== root@swamp (/dev/pts/0) /src/FreeBSD/NDIS/Realtek-8191/realtek-XP/guts 114 -> ifconfig wlan0
What does that message mean? Went looking, and it seems that there should be a file /etc/regdomain.xml with information about regulatory definitions. And I had none. It was easy enough to find one, but why wasn't it there in the first place? My guess is that it's because of a glitch in the update process: this disk was cloned from my test box, as the name swamp shows. After that I got a better result:
mcastrate appears an unfortunate choice of keyword name. But then I tried to use it, and got a subtly different message:
=== root@swamp (/dev/pts/0) /src/FreeBSD/NDIS/Realtek-8191/realtek-XP/guts 129 -> dhclient wlan0
=== root@swamp (/dev/pts/0) /src/FreeBSD/NDIS/Realtek-8191/realtek-XP/guts 130 -> ifconfig
Why that? I still don't know. How do you even investigate this kind of problem? The access point is working, because Yana and Sundance used it last weekend. It supports 802.11g, so that's not the issue. Probably the next message to look at is one I've seen in the dmesg output:
But that doesn't happen every time I create and destroy the interface, so it could be a red herring.
| Saturday, 24 July 2010 | Dereel | Images for 24 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Λκουλ: WPA for WPA's sake?
|
Topic: technology | Link here |
I'm still having difficulty getting the WLAN running on λκουλ. Somebody suggested that I needed to configure WPA, something that I don't use, but they thought it might need it anyway. So ended up with this in /etc/wpa_supplicant.conf (who thinks of these terms?):
Added the following to /etc/rc.conf:
Not that I use DHCP either, but that was the suggestion. And the result? Nothing, not even an ndis0 interface. I had to kldunload and kldload it to even get that. And wpa_cli couldn't talk to the WPA supplicant (who thinks of these terms?). Nothing I have been able to do so far has helped. I suppose I should look at the source of the Linux driver. But that doesn't work either since some recent kernel changes. I wonder why I should bother; maybe I should just return the machine.
| Sunday, 25 July 2010 | Dereel | Images for 25 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Λκουλ: stalemate
|
Topic: technology | Link here |
I can't think of anything else that I can easily do with λκουλ to get the WLAN to work. I'll put Microsoft back on it and see if I can get it to work like that. It's beginning to look as if it'll go back.
Started backing it up, which first involves creating an enormous file full of zeroes to make the disk image more compressible. On the FreeBSD partition it was about 180 GB, and it took a surprising amount of time to delete. Given my previous problems with deletion under Linux, timed it there. Surprise!
=== grog@lkoyl (/dev/pts/0) /var/tmp 10 -> ls -l foo; time rm foo
That's exactly the opposite of what I've been experiencing. I should have gone back to FreeBSD and measured things there, but it takes too long. Certainly it's an indication that there's more to this issue than I thought.
| Monday, 26 July 2010 | Dereel | Images for 26 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Λκουλ: running Microsoft
|
Topic: technology | Link here |
Started copying the Microsoft image to λκουλ today.
=== root@swamp (/dev/pts/0) /var/spool 3 -> bzcat /src/Images/LKOYL.bz2 > /dev/ad0
Then it occurred to me that that might not be optimal for the disk. Ran iostat and found:
=== root@swamp (/dev/pts/1) ~ 4 -> iostat 1
5000 transfers a second! Clearly this was into a big on-disk cache, but I was surprised that the interface could take that many transfers. Decided that it would be better to block the writes into, say, 32 kB transfers.
=== root@swamp (/dev/pts/0) /var/spool 5 -> bzcat /src/Images/LKOYL.bz2 | dd ibs=8 obs=32k of=/dev/ad0
=== root@swamp (/dev/pts/1) ~ 6 -> iostat 1
Amazing! Why should it slow down at all, let alone by so much?
Finally finished copying the image and fired it up for the first time ever. It was also the first time I used “Windows” 7. They've made the EULA a whole lot shorter now:
|
|||||||||||||||||||||||
The first boot took for ever; it seems that it does all its adaptation to the local hardware after delivery, and it must have taken about 15 minutes. When it finally said it was ready, the wireless connection didn't work—if I asked ipconfig. According to the hardware device window, it was working. Finally found a control windows similar to what Microsoft XP uses (look at the minuscule icon bottom right, which some say looks like a radiating antenna dish), and played around with it. It's not clear what was wrong, but it was quite happy to take my settings, and didn't even want to reboot afterwards.
Got hold of a firefox, but there seems to be no way to get Emacs bindings under Microsoft. I'm not sure it makes much difference: this tiny display is too small for anything useful. I might as well take it back. That'll also save me further pain with the WLAN connection.
Still, one thing is amusing, the name λκουλ. I got a message from Damon Blum with the subject line “Damon Blom akoua ports net/bwi-firmware-kmod Greg Lehey ”. So he recognized that the “Λ” was a broken “a”, but not that the “Υ” was a “Y”. Andy S (he keeps his surname secret) also opined that you can't do a Google search on “ΛΚΟΥΛ”. But you can, and they're not even all my hits. Two are mine, two contain references to my pages in the hit text, but not in the page content, and two are books in http://books.google.com/: the epistle to the Hebrews, 12:4, and Xenophōntos kyrou paideias biblia oktō . The characters recognized are nothing like “λκουλ”. The first is clearly the text “ἀκὀμι”, and the other is “δοκοῦσιν”. That's remarkably inaccurate recognition:
|
|
||||||||
|
|
||||||||
| Wednesday, 28 July 2010 | Dereel | Images for 28 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
UTF-8: how?
|
Topic: technology, opinion | Link here |
A discussion about the merits of Apples on IRC today. It's interesting, not to mention disturbing, how many people now use Apple, Microsoft and Linux as well as, and in many cases in preference to FreeBSD. Not that people are necessarily happier; few have as many problems as I do with “modern” GUIs, no matter who they come from, but they have a fair amount of criticism too. So what do we use Apples for? For things that FreeBSD doesn't do well, including hibernation. I use my Apple mainly for scanning (SANE (which I claim should stand for “Scanner Access Not Easy”) isn't really a viable alternative), and for reading SDHC cards (for some reason my version of FreeBSD doesn't recognize the reader, though that may change with an update.
Sue Blake suggested that Apple had good UTF-8 support, so went investigating that. I suppose it depends on what you want, and in this area I'm really not sure what the best approach is. There's clearly a trade-off between flexibility and ease of use. In general most people type in a specific language with a specific keyboard layout; with practice you can type at up to 8 or 10 characters per second. I can't see a way do that if you have tens of thousands of characters to choose from.
So what are the options? They appear to be:
Switch keyboard layouts. This is the oldest, the easiest to use if you want to write in only one language and don't need special characters. But beyond providing a good starting point, it doesn't really address the issues at hand.
Compose key: press or hold down a special key and then enter a number of characters which specify the resultant key. For example, the sequence Compose T H might generate the single character Þ. I've had a variant of this for decades as an Emacs macro, and only recently started to use the functionality supplied with X.
The problem with this approach is that it is relatively limited. How do you input Greek characters? Maybe it's possible, but then the second problem becomes apparent: almost total lack of documentation. I've found a number of definition files on my system, with names like /usr/local/lib/X11/locale/el_GR.UTF-8/Compose and /usr/local/lib/X11/locale/en_US.UTF-8/Compose. It's not clear which of those are used, and the man pages are no help. Following what seems to be a trend, there is no man page Compose(5) on my system, and the web-accessible man page doesn't say either. It does claim that they're in a very different place, though.
I've been here before, and once again UNIX timestamps come to my aid:
=== grog@dereel (/dev/ttypk) ~ 85 -> find /usr/local/lib/X11/locale -name Compose | xargs ls -lurt
The references to the UTF-8 variant are old and correspond to my examination of the topic four months ago. At the time I didn't realize that I was using the iso8859-1 variant. But that explains the issues I have seen: it won't allow me to compose UTF-8. But why am I using that one? Compose(5) gives the answer:
The compose file is searched for in the following order:
- If the environment variable $XCOMPOSEFILE is set, its value is used as the name of the Compose file.
- If the user's home directory has a file named .XCompose, it is used as the Compose file.
- The system provided compose file is used by mapping the locale to a compose file from the list in /usr/share/X11/locale/compose.dir.
OK, that's doable. But looking at /usr/local/lib/X11/locale/en_US.UTF-8/Compose, I don't gain much:
If I read the man page correctly, the <U1000093C> and <U10000915> are individual keys themselves, so I can't even use these combinations. Even if I could, it starts getting incredibly complicated.
An alternative would be to enter the UTF-8 keycode directly. This seems reasonable, and I'm sure I've seen it, but I haven't been able to find any way of doing it. Somebody tells me that Apple can do this if you hold down the option key and enter the code, but that doesn't work correctly for me: it does create some character, but there's one for every keystroke, and of course I can't find anything in the Apple documentation. But it's an area to investigate.
Of course, who needs a keyboard when you have a mouse? You can have a keyboard or a set of characters pop up and just select them with the mouse. That would be impossibly painful for normal text, but for entering special characters once in a while, it's bearable. Wikipedia offers this approach when writing pages, unfortunately hiding it where you can't see it and the text you're editing at the same time. Apple has this too, with one of its forgettable names: “Character palette”. Went looking for that in “Mac Help” and got all sorts of irrelevant documents. With a little help from my friends and a bit of playing around, found that I can configure to use it, along with a lot of specific keyboard layouts, and select them by clicking on—wait for it—the national flag at top right next to the other unrecognizable icons. At least it's possible to use it.
| Thursday, 29 July 2010 | Dereel | Images for 29 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
UTF-8 entry, continued
|
Topic: technology | Link here |
More investigation of UTF-8 entry on Apple today. Result: it works, once you get past the lack of documentation. First you need to enable it. You do that by the non-obvious (and initially invisible) method of setting an Input Method in the System Preferences Language/Input Menu menu, where you can also enable the Character Palette (which, I'm told, has changed its name to something else in more recent versions of Mac OS). The Input Methods look like keyboard layouts:
|
|
||||||||
You select one of them from one of the minuscule, variable icons in the top right of the screen. But somehow there seems to be a layering violation. On my machine I can select between Australian, German, Devanagari or Unixcode Hex input:
|
|
||||||||
Unicode hex input works: you hold down the key marked on my machine as alt/option and type in 4 hex digits. But which input method governs the rest of the keyboard? I don't know how to tell. Empirically it doesn't seem to be the last selected “real” keyboard layout: if I select the German keyboard, where the letters y and z are reversed, and then select Unicode input, it reverts to the non-German layout. What a mess this stuff is!
|
ALDI DVB-T tuner
|
Topic: multimedia, technology | Link here |
Yvonne came back from shopping today with a 1 TB disk and a USB DVB-T tuner from ALDI. I don't know why I wanted the tuner: it cost $50, and I've already established that the Winfast DTV Dongle Gold ($34) works. Tried this one out. It didn't, and I can't be bothered to investigate further. Isn't it nice that ALDI takes things back with no questions asked?
| Friday, 30 July 2010 | Dereel | Images for 30 July 2010 |
| Top of page | ||
| previous day | ||
| next day | ||
| last day |
|
Slow deletes under Linux
|
Topic: technology | Link here |
Mail from Daryl Tester today discussing slow deletes under Linux. He suspects it has something to do with file fragmentation, and it looks as if he is right. He points me at a program called filefrag, which shows the number of fragments in a file. Tried running it against a number of files that I wanted to delete and compared the deletion times:
| fragments | size | real | user | sys | ||||
| (s) | (s) | (s) | ||||||
| 5 | 1374941948 | 0.032 | 0.000 | 0.004 | ||||
| 7 | 1478355296 | 0.031 | 0.000 | 0.000 | ||||
| 1688 | 2770630260 | 3.311 | 0.000 | 0.036 | ||||
| 3023 | 1993571176 | 2.868 | 0.000 | 0.056 | ||||
| 1281 | 2719592208 | 1.300 | 0.000 | 0.012 | ||||
| 949 | 3203344032 | 1.242 | 0.000 | 0.008 | ||||
| 1470 | 2853799768 | 3.112 | 0.000 | 0.028 | ||||
| 95 | 6991753652 | 0.117 | 0.000 | 0.004 | ||||
| 1607 | 3900287856 | 2.857 | 0.000 | 0.012 | ||||
| 3711 | 4198226308 | 9.853 | 0.000 | 0.136 | ||||
| 1560 | 2349734920 | 3.431 | 0.000 | 0.028 | ||||
| 2220 | 1471019348 | 4.807 | 0.004 | 0.020 | ||||
| 1888 | 2851456536 | 5.033 | 0.000 | 0.064 | ||||
| 2437 | 3559201148 | 4.454 | 0.000 | 0.028 | ||||
| 22 | 4886416288 | 0.069 | 0.000 | 0.000 | ||||
| 3650 | 4118056528 | 7.592 | 0.000 | 0.096 | ||||
| 13 | 10835008324 | 0.024 | 0.000 | 0.000 | ||||
| 3046 | 3636609772 | 7.508 | 0.000 | 0.072 | ||||
| 52 | 5344228060 | 0.009 | 0.000 | 0.000 | ||||
| 4888 | 6629422004 | 10.181 | 0.000 | 0.100 | ||||
| 2794 | 6545229964 | 4.472 | 0.000 | 0.036 | ||||
| 6061 | 11617944664 | 7.467 | 0.000 | 0.060 | ||||
| 14 | 6181893268 | 0.014 | 0.000 | 0.000 | ||||
These files are MPEG-2 transport streams written by MythTV. Why are there such wildly different numbers of fragments? Why, indeed, should there ever be that many fragments? Is this an indication of poor storage allocation? I thought that possibly filefrag (written by Ted T'so) might be ext3 centric, and that it might not apply to XFS, but these values strongly suggest that it does.
| Saturday, 31 July 2010 | Dereel | Images for 31 July 2010 |
| Top of page | ||
| previous day |
|
More NFS locking issues
|
Topic: technology | Link here |
Yvonne reported a read-only mailbox again this morning. I've been through that before, and I was able to get things going again with:
=== root@lagoon (/dev/pts/3) ~ 9 -> /etc/rc.d/statd restart
=== root@lagoon (/dev/pts/3) ~ 10 -> /etc/rc.d/lockd restart
But why does it happen? It's got to be a race condition, and that's particularly difficult to catch.
| Top of page | Previous month | Greg's home page | Today's diary entry | Next month | Greg's photos | Copyright information |