Thomas Legg's Geek Blog

So something in the webserver config has blown up since about 5:30pm last night HKT (5:30am EDT for the Americans). Since then pages have been sending blank headers to the fastcgi server and errors have been prolific. Bloody hell! Working on fixing this issue. Sorry for the crap.

UPDATE: 1:30pm I found a php-cgi.core file and fired up gdb. Looked like a problem with the filter.so for php, so I commented that out from extensions.ini. Everything seems far stabler from that point on.

So one huge advantage of having an Atom based netbook (BenQ Joybook Lite U101) has been having a FreeBSD testbed to play around with various technologies before trying to deploy them here.

Currently www.the-eleven.com runs on the venerable Apache webserver using the venerable prefork mpm. It also is using mod_php. Nobody is going to accuse this combination of being svelte, but it is dependable and RAM is cheap these days.

So I've been looking at nginx and spawn-fcgi (from the lighttpd project). With apache the rewrites for web software can be placed in htaccess files and run from there. With nginx, all of the rewrites have to be placed in the configuration files. This can get to be a pain quickly if you have webapps in sub-folders for other web apps.

So as a quick and dirty compromise I've been testing Apache using the worker mpm with mod_fastcgi and php built for fastcgi. Initial tests with apache bench seem to suggest I can get similar performance to the existing setup while using far fewer resources.

So you may notice some flakiness in the site over the next few days as I play with doing a full switchover to this newer model and seeing if it stands up under real world battering by spambots.

And from there I can explore the nginx configuration issues and test to see if that increases output with even fewer resources.

UPDATE: 11:11pm
So I've replaced mod_php with mod_fastcgi, but still running apache with the pre-fork model. I've already noticed a few apps failing with the current config. Hopefully I can update some of those, but they aren't mission critical, so they can wait until tomorrow. If you bump in to problems, like cgi failures, let me know in comments.

UPDATE 1:35p 24 June 2009
So I've done the cutover to the worker mpm for apache httpd instead of the prefork mpm and the change in the apache bench marks are amazing. And the server is using fewer resources.

Today's pet peeve is the stupidity of Apple's Installer packages. Stupid? Yes, stupid. I keep applications on a second partition away from the System installation. Much like I have separate root ( / ) and app ( /usr ) partitions on my FreeBSD system. Since by default, Mac OS X uses the System disk for swap space, this means it's best to keep as much free space on that partition as possible if you intend to do heavy lifting with your Mac (like compiling Songbird and other apps) and hit swap. So with the applications on another partition, I put a symbolic link between /Applications and the directory on the other partition.

Apple's installer doesn't like the symlink for /Applications and overwrites it. Then it tries to update the applications found inside the obviously empty /Applications directory it just created. For something like the Safari 4.0 upgrade that means that nothing gets written to the disk in the /Applications directory. So I have to go back and manually copy Safari back to the newly created empty /Applications, re-do the installation, reboot again and then move Safari back to the other partition, delete /Applications and recreate the symlink. ICK!

Could Apple make their installer smart enough to recognise the symlink and just work? Yes. Will they? Doubtful.

Songbird 1.1.3 macosx-ppc dmg
md5 for dmg file

I'm doubting that it'll run on 10.4. I haven't booted back to check, but not hopeful. Tested locally on 10.5, but I've got all of the build libraries, so I'm not a good guinea pig to see whether the build is self-contained.

For those of you who don't have don't have identi.ca accounts and don't follow me there, about two weeks I posted this dent.

Talked to an SE today from a company producing ARM and MIPS netbooks on use of linux. They had demo version of one running Android interface

SE is sales engineer and the company was Skytone. People asked questions about the ARM and MIPS based netbooks and I spent some time answering their questions about the products I saw at the two Spring Electronics shows here in HK.

One of the bots that I follow on identi.ca posts tech news snippets and their is an author at ComputerWorld, Seth Weintraub, who writes about such items. One of these stories was about Microsoft's plans to cripple the number of programs that can run at the same time on Windows 7 starter edition and how this might open up the ability of Google to push Android on netbooks. The first comment on that article was a tip about the Android netbook I'd seen with a link to the Skytone product page for the 680.

Mr Weintraub took that tip and posted this story on the Skytone 680 originally without even so much as a tip of the hat to me for my well-placed comment. He also took another dent of mine out of context and said it would be $100 (later rewritten to be $100-$200 after I posted my comment questioning the original $100 quotation. They also added the tip of the hat and a link to my identi.ca page.). This article opened up the floodgates for coverage on the Skytone 680.

From the identi.ca perspective it came full circle when one of the folks I followed dented about this Slashdot coverage of the Skytone 680 alerting us all to this nifty Android netbook.