The new guts for my home Linux server showed up. I read the manuals (for the motherboard and case), wrote out a 31-step process for migrating the hardware over, and with only one hiccup that lasted about 15 minutes, it went smoothly. The new server is up and running. I think it took about three hours from start to finish which was mostly spent doing mechanical stuff (unscrewing things from the old case and screwing them in to the new case, attaching and routing cables, etc.). I'm so psyched. It works! Step 1 was to burn a DVD-R of the latest backup on the old server. I tried to compress the backup directory so it'd take less time to copy to my PowerBook for burning, and the CPU very nearly overheated. So sad. Trying to back up the old server so I can replace it because it overheats, it almost overheats. So, I copied the tar file uncompressed, and compressed it on the PB.
I've put the new server through a few tests and it's been fine. I ran cpuburn for a couple of minutes but chickened out because I don't have the temperature sensor libraries installed yet. Those would let me watch the CPU temperature as I ran cpuburn, so I could see if it was gonna overheat or not. Ideally I'd be able to run cpuburn for a good hour and then rest assured that it won't freak out later on. Actually I'm not sure that I can install the temperature sensor stuff without a kernel upgrade but that's something I'll have to figure out sooner or later. Anyway, the new server works and I can't wait to benchmark it on various stuff like building the application I'm working on, um, at work, and running all the tests which are mostly CPU-bound.
Last night I finally finished researching the purchase of the guts for my new Linux server at home. I had originally wanted to turn my no-longer-needed PC desktop (a 500MHz Pentium 3 based system) into the new server, allowing me to slowly migrate services off of the old server, so that there was no hurry and I could work on it during little lulls in other stuff I was doing. (Having a very detailed migration to-do list helps with this sort of project, so it doesn't take very long to remember what you were working on last time.) Unfortunately the new server kept having cooling problems. I bought new fans for the case, new fans for the drives, a new heatsink and fan for the CPU, and it just wouldn't behave. I thought I had it fixed with the new CPU fan, but when I was trying to run Bonnie on it recently, it kept overheating and shutting down. So I decided to quit throwing money (and time) down a hole and just buy something new and healthy, with a warranty and stuff. (I bought the old hardware for $100 at a Sapient post-layoff fire sale back when I worked there, which was over 3 years ago.)
Well, if I hadn't bought so much new hardware recently to try and beef up this machine I probably would have Just Bought a Dell, Dude. I decided to see how hard it was to spec out the new guts myself.
Well, it's hard. The main problem is that the market for PC parts is so huge that it takes a lot of time just to understand the main product lines and focus on a few products. The sales doubletalk is in no way helpful. Did you want a high-performance economy system, or an affordable high-performance system? We have 1500 models to suit every need! They're all an equally good deal. Don't you know what you want?
Fortunately there are all sorts of magazines and web sites to help you. They come in two categories:
Some of them have incredibly useful tables like this one. Thank god it's all laid out so clearly. I mean, the one I should buy is just jumping right off the page? Can't you see it? Cross your eyes and look for it, it's shaped kinda like a sailboat. Keep looking...
The problem is that there are too many people who just love to watch the PC industry and keep up with every detail and maybe buy something every so often when they see a good deal appear. These people are called "Enthusiasts" by the industry. Playing around with their computer without actually doing anything is the whole hobby. Several years ago they figured out that they could cut holes in the side, put LEDs in the computer, and paint it all funky, and their computer nerd friends would be impressed. Some of them would mess with their PC's hardware to run it faster than it was supposed to be run, and then buy extra powerful fans (or far more complicated cooling systems). They're called "Overclockers". Picture a Daihatsu with a giant spoiler and nitrous oxide. And then what do they do with these super-tweaked computers? Play games. People who actually need fast computers to do work would just buy a faster computer instead of spending all that time putzing around under the hood.
It's challenging to get past the web sites and magazines that cater to these people and to the more useful facts like where the sweet spot in price/performance is right now, rather than the latest-greatest high-end hardware that nobody in their right mind would buy because it's wildly overpriced right now. I suspect that this is because Dell keeps pressure on this market segment; if it takes days to figure out what to buy and Dell's system is on sale for about the same price (and is 2-3% slower than what you built yourself, maybe) and you could have just ordered it and it'd be here by now, and working, what's the point?
Well, some people worry about Buyer's Remorse. That's the phenomenon that happens when you educate yourself, buy it, take it home and decide you love it and are very happy, and then the next week a new version that's 5% faster and 10% cheaper comes out. This is a problem that has existed for at least 20 years. I've been nailed by it several times. That's one reason why sane people with better things to do (a category which hopefully includes me) worry and waffle and try to make sure they know what's about to happen and waste time reading all of these damned web sites that are really just for gamers. But the real solution is to pick a trustworthy vendor, buy something with the information you have today, and then stop looking.
Fortunately, I found that Computer Shopper magazine (yeah, the paper kind) had very practical, what-to-buy-today sort of information. They had a review called "Pick the Right PC" which was for whole systems, but I was able to see what processors and motherboards the recommended systems had in them, which helped me.
That, combined with some useful "basic info" articles in PC Modder magazine (a magazine intended for the hobbyist crowd) helped me figure out what I wanted to get. I even found out that there was a particular sorta-high-end processor that had been recently introduced at a good price. Here's what I ordered:
I started with the processor. AMD is giving Intel quite a fight and I want to support AMD because they aren't trying dirty proprietary tactics like Intel did with their new 64-bit instruction set. In other words, AMD is willing to compete openly and Intel isn't, and AMD is winning anyway. I searched around for info on their processors and then said "duh" went to their own web site, which was very informative. I looked at the Opteron processor, which is for servers, but the main reason to get an Opteron instead of an Athlon is if you want to have multiple processors, and when I added up the 2 processors and motherboard and RAM the price was out of my price range (the cheapest 2-Opteron combo with 1GB RAM was about $1000). The Athlon 64 comes in a couple of forms, one of which has twice the CPU-to-memory bandwidth as the other at the same clock rate. The nice thing is that they just came out with a slightly slower version of the dual-data-path one (the better one) that's about half the price of the cheapest one that was previously available. That cheaper one is the 3500+, compared to the 3800+. Also, the same motherboard and socket ("Socket 939") will fit the even more expensive fastest Athlon there is, which is the FX-53. So, hopefully there is some upgradeability available to me since I can just buy a faster socket 939 CPU later on if this quite fast Athlon 64 3500+ starts to seem slow to me. :)
I found a nice article about the Athlon 64 3500+ that helped me make this decision. First, the chart on that page says a lot. It's supposed to be there to help benchmark readers figure out how they tested, but it also helps me understand what a "good" configuration looks like. (They wouldn't want to put together a handicapped system and then get lots of angry reader mail about it.) So, there's a hint in that chart about what motherboard that reviewer thought was an appropriate match. Of course, the benchmarks are more useful. Pretty much across the board, the Opteron 150 and Athlon FX-53 beat up the Athlon 64 3500+, and the Athlon 3400+ is close to the 3500+. That's OK with me though because of the price differences.
An Opteron 150 (note: not dual-CPU capable), a Tyan S2850G2N motherboard, and 1GB of registered RAM cost about $950. If you want to be able to upgrade to a dual-CPU system then you'd need an Opteron 250 and an ARIMA HDAMB motherboard plus 1GB of registered RAM, which is over $1400, and you'd still need to shell out over $800 when you wanted that second processor later.
The Athlon FX-53 uses the same socket 939 motherboard as the Athlon 64 3500+, and the same RAM, but costs over $450 more (almost two and a half times as much as the Athlon 64 3500+). Look at those benchmarks... not worth 2 1/2 times as much. For most of the benchmarks it would be a better deal to just get the Athlon 64 3800+ (which is priced just about halfway between the 3500+ and the FX-53).
So, it really looks to me like the 3500+ is where the CPU sweet spot is. It's fast, but not the fastest, and still it's pin-compatible with the 3800+ and FX-53 (and future chips in AMD's processor roadmap) so a whole new motherboard/RAM/CPU combo isn't needed when I feel like upgrading later.
The motherboard was a tough choice. Because I'm going to run Linux on this machine instead of Windows, the choice was easier to make. nVidia makes the chipset in some of the socket 939 motherboards, and they apparently aren't very Linux friendly. So, I stopped looking at mobos with an nVidia chipset. VIA makes a socket 939 compatible chipset and according to some folks on the various Linux newsgroups (thank you groups.google.com) they are nicer. So, it was down to the ASUS A8V Deluxe and the Abit AV8 motherboards. Basically they have the same features (almost), same chipset, and cost about $125 +/- $10. (It is pretty stupid that the companies are named ASUS and ABIT and their competing products are named AV8 and A8V, by the way. I've never been so aware of the value of a decent branding firm and a marketing department before now.) I found a number of messages on forums and newsgroups advising folks not to deal with Asus, culminating in Asus Anti-Linux Attitude Sucks. OK, so Abit it is, then. I did some more searching just to find people who had used this stuff and found a few resources, but apparently this hardware is so new that there just aren't that many people using it yet. I'm taking a bit of a leap but since the vendor is well-liked by Linux folk (apparently) and the chipset is the same as a motherboard that is known to work, chances are good that it'll be OK, and that any hiccups won't be met with an icy stare from the manufacturer.
The RAM was a pretty simple pick. The company I ordered the CPU/motherboard combo from, Monarch Computer, has a configurator that lists parts that will work. Abit's web site has a Compatible Memory page and Corsair also has a memory page with links to a buyer's guide PDF that all agreed: this is the part that Corsair sells that will work in this setup. The price wasn't too high so I didn't spend a lot of time shopping around. Mainly I just want it all to work when I get it, and if I left $50 or even $100 on the table that I could have spend a whole day trying to save, that's fine because my time is worth a lot more than that.
In case you're asking yourself "hey Mr. Mac owner why didn't you look at a Mac too?", the reason is that I want to run Linux to stay in touch with what's going on in that world. I know you can run Linux on a Power Mac G5, but a secondary problem is that the G5 comes bundled with a bunch of stuff I don't want to pay for (case, power supply, drives, video card) that add up to quite a bit. A new dual G5 would be more than I need for more than I want to pay, and a used one wouldn't have a warranty and would still make me pay for that same bunch of items I already have. I think if I were going to Just Buy a Dell, I would have looked at used G5s on eBay as well. Likewise, if I were in the market for a dual Opteron server, I'd look at a G5 too. (The Xserve has too much of a price premium attached for compactness, which isn't something I care about in a server right now.) But for now I want to keep one foot in the Linux-on-PC camp because that's the bleeding edge, and I'm willing to do some bleeding to keep up with the latest developments there.
On a related note, Grady Booch is a switcher.
I wish I didn't have to care enough about FireWire enclosures and chipsets to need to read stuff like "Utilizing High Capacity ATA-6 (>137GB) Drives in FireWire Enclosures". But it's very informative anyway. Conclusion: I need a new FireWire enclosure that can handle big disks, 'cuz the one I have won't. Dang.
I also wish that replacing this small business firewall/router with this small business firewall/router wasn't necessary to solve a problem that shouldn't have existed in the first place. I've never seen a network appliance reboot itself after a couple of minutes of uptime, several dozen times a day, especially when it's the second identical unit I've tried. Well, that's the problem with the "appliance" approach to complex network gear - if it doesn't work, too bad. Perhaps I could have spent a bunch of time debugging the Linksys gizmo, since I hear they're based on Linux and that there are third-party open source firmware updates available. But, there were 3 things that made me not want to go that route: (1) The router is shared between my employer and another company in the same office, so urgency was high. (2) I don't want to volunteer to be the guy who gets called every single time that router does anything wrong, or might seem to be the source of a problem to someone who isn't a networking expert. Hey, I can't get to this web site; call Jamie and ask him to see if the router needs an update or something. I'm not looking for that sort of job security, nosiree. (3) The cheapest solution was just to try and replace the thing that was misbehaving, once I had troubleshot the situation enough to know that it was definitely where the trouble was. I think it cost something like $80, and the price of my time multiplied by the time it would have taken to hack some unsupported open source firmware alternative just wasn't worth it. This is why people buy the $80 gizmo instead of setting up a super complicated PC based solution - 90% of the time it works, and it takes 10 minutes to set up, and if it breaks, you throw it away and buy a new one. The alternative, which is usually "I have this old 486 that's not doing anything" tends to take a huge amount of time and effort to set up, and then it uses a bunch of power and space and every moving part in it is likely to break pretty soon.
This situation makes me think about embedded Linux in the abstract. I don't have the source to the Linksys router firmware, nor do I want it. I don't want to keep it around, I don't want to subscribe to some mailing list so that I can stay on top of the latest changes to it. I want somebody else to deal with it. I'm vaguely glad that it's Linux based so I have some sort of assurance that gaping security holes will be fixed even if Linksys doesn't have the budget to track them all down itself, though. The value added by Linksys is not that they paid someone to write a bunch of firmware; what matters is that they take something immensely complex and configurable, figure out how to make it do just a limited set of things that I need, put it in a tidy little gizmo, put a web UI on it, write documentation, and sell it for $80 or less. As it happens, the "90% of the time it works" wasn't enough, but the fact that there was another gizmo for the same price that did work translates to something more like "99% of the time, one of them will work" for about $160 if you buy the wrong one first. That's still great. I am quite curious as to who wrote the firmware for the Netgear firewall, though - is it Linux too, or something else open-source, or proprietary? It's just one data point but it would be funny if Linux failed where proprietary code worked. That's orthogonal to the matter of buying commodity appliances instead of the false economy of putting together cheap/free hardware with free software and using extremely expensive labor to make it all work.
Hey, Microsoft, Nice mouse! Where's the bluetooth version? Oh yeah... they just decided to give up their anti-Bluetooth stance with XP SP2, so I guess we'll have to wait.
It's apparently possible to patent a peanut butter and jelly sandwich. Actually the patent is for a particular kind of sealed PB&J sammich, if you read it a bit.
Fuck You and Your H2 is pretty darn funny. It's scary how many user-donated pictures there are on the site.
When subwoofers just aren't bassy enough, there's The Buttkicker.
John Kerry made a pretty darn good speech on the situation in Iraq.