|YaK:: WebLog #535 Topic : 2004-09-03 06.50.52 burddell : the linux experience 2 years later, and infoworld||[Changes] [Calendar] [Search] [Index] [PhotoTags]|
|[Back to weblog: pretention]|
<a href="http://www.offramp.org/~ghuang/">Geoff</a> wanted to get a new machine and I convinced him to get an <a href="http://www.shuttle.com/hq/product/barebone/default.asp?B_id=25">Athlon64 Shuttle</a> (and a GeforceFX 5700 video card). He wanted to try linux on it, and since he uses KDE, <a href="http://www.suse.com">SuSe</a> was his first choice.
SuSe sells an AMD64 version of their 9.0 distribution for about $120 ($30 more than the 32-bit version). We tried to find it at Fry's, etc but no one had it. So we downloaded the AMD64 net install ISO and booted.
First thing we learned is that you must flash the BIOS before you do anything. Every forum and linux site on this subject says there the BIOSes were/are buggy as hell, and any problems can probably be resolved that way. So, flashed the BIOS and we were on our way with SuSe. Until...
SuSe's net installation simply did not work. FTP, HTTP, you name it. Didn't work. At all. Web search turned up that many people had this problem even on 32-bit. Way to go, SuSe. Geoff was discouraged, but he really wanted SuSe so he could get paid support, etc. So he took into work the next day and somehow got SuSe installed on it.
Next, we tried to run X. It didn't work. The screen would blank, or we would get weird fuzziness on the screen. I learned some time ago that when you have a problem with an app (especially opensource), go check the ChangeLog for changes in the area you are having your problem. So I went and look at the linux kernel changelogs and sure enoughh there was a fix in the 2.6.4rc1 kernel for nforce3 AGP. Great. So I got the source and went to recompile it, when...
I decided to make sure the gcc was a sane version that didn't have AMD64 issues. It had gcc 3.3.1, which is quite a few months old at this point, and there were some significant AMD64 changes/fixes in the releases since then. So, I decided to bootstrap a new gcc 3.3.3 before compiling the kernel. This was actually easy (and finished in about 90 minutes), which is a several order of magnitude improvement over my experience building gcc a few years ago. gcc built, and the failed tests were expected, so I went to build the kernel.
Building the kernel is easier now too. I had never dealt with grub before (boot loader replacement for lilo, lilo won't even work on 64-bit supposedly), but it was pretty easy to deal with. Got the kernel configured, compiled, rebooted (a couple of times because I forgot some things), and then re-ran X. Slightly different problems, different kind of fuzy screen, but hey -- an improvement.
We were having a <a href="http://www.yak.net">Yak's Headroom</a> at my house that night, so I got some troubleshooting help from some of the clan. We screwed with the X config, and it turned out that was the wrong file and yast (SuSe's config tool) was reading/writing from something else entirely. It kept setting the monitor to some weird refresh rate.
So I went and looked at the recent XFree86 changelog. Sure enough, 4.3.99 RC3 (soon to be the 4.4 release) had a fix SPECIFICALLY for the GeForce FX 5700. Fuck me in the goat ass. I was going to have to rebuild X. (Which in turn meant rebuilding KDE and got knows what else against that version of X.) Son of a bitch.
At this point, <a href="http://www.appelbaum.net/">Jake</a> said that if we had to rebuild the world, we migt as well use gentoo. I had heard people talk about gentoo in religious tones, so I assumed it was a bullshit linux groupie thing that was going to suck as much as this but have the support of a bunch of self-righteous fuckwits. I couldn't have been more wrong. We decided to try gentoo.
gentoo just fucking worked. praise jesus and <A HREF="http://www.amazon.com/exec/obidos/ASIN/B00009V7P8/matthargettbl-20">say hallulujah, if it means something to ya</a>. They had an AMD64 "liveCD" which booted into a fully functional linux. gentoo is source based (binary packages are available generally as well), so it really optimized the compile the world yourself experience. It kind of reminded me of Slackware '96 (the first distro I used at <a href="http://www.cityscape.net">CityScape</a>), but unlike Slackware and RPM, it had dependency resolution that just worked. I was very impressed, which is a sad thing considering some retards think linux will overtake the desktop soon.
I think that makes sense for "managed" desktops where IT is in control of everything, users don't have admin/root access, etc. But as for the average Suzy Q., she won't put up with this bullshit. And she shouldn't.
That being said, gentoo has the best documentation for an install of any linux distro I ever used. (Slackware, Redhat, Debian, Turbolinux (where I worked for a year), SuSe, ...) You were building the world, but the step by step instructions were very clear, comprehensive, and made everything as easy as it could be. (Read above "which is a sad thing".) That, and their IRC channels and forums actually have helpful/nice people! Other than some discussion about the X license change, the normal retardation I associate with linux groupies was absent. Awesome.
Having spent 2 days on SuSe, we got gentoo up and going in about a day of Geoff and I tagteaming on it. gentoo even had a package of XFree 4.3.99 RC2, right before they made their license change. Everything just worked, including KDE, KDevelop, KOffice, mozilla, etc. (KOffice took the longest.) One of the benefits of building the world is seeing the warnings go by (if you are a fuxing l0ser like myself). Seeing "pointer cast to int of a different size" is not a warning you want to see when compiling on a 64-bit platform (hint to the mozilla folx there). Unfortunately, <a href="http://valgrind.kde.org/">valgrind</a> doesn't work on x86-64 (yet).
Speaking of <a href="http://www.mozilla.org">mozilla</a>, the mozilla foundation has moved there office just down the street from my house. Maybe I should go bang on their door and give them some access to this machine so they can fix their stuff. Mozilla appears to run fine, but whereever those 64-bit pointers are being truncated down to 32-bit (int), there's gonna be some problems. I'll bet I can detect that kind of thing in the binary with BugScan really easily.
Speaking of BugScan, the <a href="http://www.infoworld.com/article/04/02/27/09TCbugscan_1.html">InfoWorld review</a> finally got published. (We sent them a box in September, still haven't gotten it I don't think.) The review is pretty high level, but I'm glad to get a good score. He thought the price was a bit high, but we keep hearing from our paying customers that for what it does, we aren't charging enough. He mentioned the new functionality of 2003A in a sidebar which appears on the website, but not the print article. Oh well, hopefully it generates some more revenue for us.
Speaking of BugScan revenue, I came to a realization recently. Originally, I thought funded development of features would help jumpstart the company so we could hire more people and do some things that are a little more ambitious. (I'll be talking about one of them at Blackhat Vegas 2004 if I get accepted.) People like the idea, understand that we're self-funded, but they ultimately don't want to work that way. It usually turns out they want us to do some work to implement features, and then they don't have the money anyways. We have pretty consistent sales, just not enough to do a whole lot with other than pay me a minimal salary so I can keep up with the mortgage and pay a junior programmer. Don't get me wrong -- consistent sales are a great sign we're doing something right in our current market. So, to jumpstart the company we're going to start looking for a little angel investment.
I'm re-reading <A HREF="http://www.amazon.com/exec/obidos/ASIN/0060517123/matthargettbl-20">Crossing the Chasm</a> and <A HREF="http://www.amazon.com/exec/obidos/ASIN/0787952184/matthargettbl-20">High Tech Startup</A> to get back into the mindset I was in last year when we started the company. I also read <A HREF="http://www.amazon.com/exec/obidos/ASIN/0066620996/matthargettbl-20">Good to Great</a> last year, and it was excellent, but I got it's concepts the first time around. I think I'll skip <A HREF="http://www.amazon.com/exec/obidos/ASIN/0060521996/matthargettbl-20">The Innovator's Dilemma</a> this time, which was interesting but extremely dry and long-winded (in my opinion).
<code> </ramble> </code>
|(last modified 2004-09-15) [Login]|