Slashdot Mirror
US Army Needs Linux Workstation Advice
Shawn wrote in asking for suggestions on high-end Linux Workstation configurations for the US Army. Aparently, they are looking at replacing some SGIs with cheaper PCs, and this being Slashdot, I'm sure some of you might be able to make some improvements to their suggested configuration. I can't tell you how glad I am to see the US Army giving Linux a chance. The US Navy could probably learn a thing or two from this. Click below for the full text.
Shawn asks: "The US Army is looking for advice on purchasing several high-end Linux workstations. These systems will be used in a post-flight processing environment. We routinely process raw data measurements in excess of 1 million samples/second. This processing can utilize several GB of disk space and up to and exceeding 1GB of RAM. We are currently using SGI Octane systems but need to switch to the lower cost PC environment. This will allow less costly while still providing excellent performance. Let us know any advice you may have or insights we may not have thought of.
Here are the specifications we are currently considering:
Motherboard
- Intel® 840 chipset
- RIMM slots for at least 1GB dual-channel Rambus® memory
- Integrated dual ATA-66 EIDE controllers
- Ports: USB (2), serial (2), Parallel, PS/2
- Integrated Ultra 160/M SCSI and Ultra/Wide SCSI Controllers
- NO INTEGRATED VIDEO
- Intel® Pentium® III 733Mhz (slot 1)
- .18-micron Coppermine architecture
- 256K cache
- 133MHz FSB
- 1GB Rambus® memory
- Full tower
- 300W or greater power supply
- 4 or more 5.25" exposed drive bays
- 1 or more 3.5" exposed drive bay(s)
- Two or more small fans blowing air directly across the drive bays
- 9GB Ultra 160/M SCSI (7200 rpm) hard drive (internal)
- 36GB (or larger) Ultra 160/M SCSI (10000 rpm) hard drive (internal)
- LS-120 internal superfloppy (IDE)
- 5X (or faster) DVD drive (internal - IDE or SCSI)
- Diamond Viper V770 Ultra 32 (AGP 2X/4X)
- Viewsonic P815 21" monitor
- 3Com 3C905B-TX ethernet card (PCI)
- SoundBlaster PCI128 (PCI)
- Powered speakers with wall adapter
- Soft-touch keyboard (no keyclick)
- Microsoft Intellimouse
- Red Hat Linux 6.x
- 3 year parts and labor
Thanks,
US Army Test Facility."
Actually, their business is to keep other people from killing me, or forcing me to do something I
don't want. Something that I heartily agree with.
Addtronics 6890a
And personally, I LOVE this sucker. Nice and open, even with tons of stuff in it, and it can take about 5 extra fans. Oh, and the Mo-Bo tray slides out with all the expansion cards in place with 4 screws removed.
-- IANAEG - I am not an elder god.
Let's see if VA Linux can handle this.
Hates people who have stupid little sigs
Avoid the Intel CPU's. The AMD K7's outperform the Intel Pentiums, anyway.
Since this is a government bid, presumably security is a concern. The problem with the Pentiums is that the CPU microcode can be changed
on you, and you'd never know it.
This problem gets even worse with the Merced/Itanium/whatever-the-64-bit-chip is called.
With the embedded crypto, changeable microcode, and the moves to integrate ethernet on the motherboard, the 64-bit CPU will be wiretap ready.
This is presumably why the Chinese have moved away from the Pentiums, and Europe appears to be heading in the same direction.
The K7's don't have this problem. Nor do the PPC, or the DEC Alphas.
Sometimes this kind of stuff just irks me - how clueless are people anyway??
This is a DATA PROCESSING machine. You do NOT need a 128bit sound card plus massive speakers. If any kind of sound is going to come out of this thing, an average SB16 or AWE32 is more than adequate.
You also do NOT need a Viper 770 video card. What the heck are you going to display on your screen that you would need 32 Mb of video RAM for?? Just an ordinary video will do - 8Mb at the most. This is X-Windows, not Quake.
No need to specify UltraDMA66 controllers if the only thing you're going to hang off it is a DVD drive.
The Athalons will still give you a few percent higher performance on math than the PIIIs, you may want to go with that instead - though you may not have much pick of motherboards.
Other than that, looks like a darn nice set of specs.
If it is specialized, I would recommend a simple window manager such as BlackBox or IceWM. There is no need to burden your system with unnecessary tools and applications. If, on the other hand, it is going to be a user friendly system, I'd have to recommend using KDE or GNOME, giving preference to your personal aesthetics. The added complexity of GNOME or KDE should be justified.
In terms of hardware, I must point out that USB support will not be available until kernel 2.4, IIRC, but that will be arriving soon. Also included will be firewire support, again IIRC, which should encourage you to investigate using that as a possible alternative to SCSI. Again, it is application dependent, and also subject to some personal preference.
In terms of video, the Diamond Viper is a bit of overkill if you'll be using BlackBox or IceWM, and you could save yourself a bit of money by getting something a bit less hefty, and possibly including TV support, such as the ATI Rage128. Since the TNT2 chipset is being backed by NVidia under Linux, it is a decent choice, IMHO.
With respect to the processor of choice, I see no real reason not to get an AMD instead of an Intel if the price is right and the speed is sensible. Alternatively, with Linux, you can go completely Alpha, or Sparc (although I would avoid Sparc, but that's a personal aesthetic thing), and not have to really worry about interoperability with other platforms given the nature of Linux.
Well, I hope that helps.
The first three companies are pure Linux. The fourth is getting good advice on Linux systems from people in the community.
You absolutely want as little variation on your hardware as possible with these machines. This will up the price of each machine. However, it will also make them easier to maintain, repair, upgrade and find drivers for. The additional expense pays off in the long run, if you are going to deploy thousands of these things. Make this a condition of your contract.
Personally, I say go with ASL, if only because they're nice guys.
NB- I don't work for any of these companies, even (as far as I know) indirectly.
Finding God in a Dog
I have a feeling this sounds like a storage machine - means you should have a good amount of ram, but regular PC100 or PC133 will work fine. Dual processors? maybe, if you are doing stripeing or mirroring with your disks.
Also don't limit yourself to RedHat - remember redhat is not linux - you can go ahead and look at debian, stormix, mandrake, suse, and the rest for a distribution that fits your needs, or one that just spikes your interest. (I personally like stormix)
if you can get past the windows mentions, look at the god-box and the other boxes over at Ars Technica. They have some great reccommendations on good value hardware.
Lastly, i saw a few complaints about the 3c905b cards in earlier posts. We religiously use these cards (albiet in windows NT machines) here at work and have had nothing but good experiences with them. I love the 3com and i have a hard time using anything else (execpt if it is free of course :-> )
--onyx--
Sir. We at the NSA are not aware that we have any
such sense of humor.
This is a difficult argument to counter and one I have been dealing with for a long time. As a West Point graduate, I am often asked why I chose to go in the Army. My answer is usually, "Too complex a question for the time alloted for the answer." However, one of the major deciding factors was studying about Mei Lai in high school. THe majority of people in a South Vietnam village were massacred by US troops. Only one person went to prison: The Platoon Leader. I remembered thinking, "If I had been in his position, that would not have happened." That is why a free thinking computer geek from liberal Massachusetts decided to forgo the college (read party) career he had been looking forward to for most of his life to go through the rigors of military schooling and traiing. I graduated and was an Infantry officer (not a very good one, in retrospect).
After vietnam, the Army was viewed as a dishonorable profession. As a result the army of the 1970's was riddled with problems. It took many years and a shift to a president concerned about the military to fix them.
The fact remains that we will have a military. While it would be wonderful if we could ensure world peace without guns and bombs, it is not going to happen soon. Work towards it, it is a wonderful goal. But understand that in the interim there will be men and women standing guard. If we the American public dedicide to treat the profession as honorable, then honorable men and women will be drawn to it. It is not a fixed thing; the attitude we spread today will directly affect the actions of the military tomorrow. Express your concern about the choices the civlian leaders make when they decide to deploy our troops. But support them men and women who have to make very difficult decisions. In short, ensure the morality of the military lines up with the morality of the people. The Military of the past has made mistakes. It is up to us, all of us, to learn from them.
Adam Young,
1LT, Infantry
(USAR)
Open Source Identity Management: FreeIPA.org
According to Tom ddram outperforms rambus anyway, due to rambus's inheritantly worse latency. Also if reliability is a concern :-) keep in mind that a slew of Rambus-related problems have dogged the new rambus motherboads/chipsets.
Life's a bitch but somebody's gotta do it.
Point by point, dissection of the specs.
1. i840: The best choice of chipset. No complaints.
2. dual ATA-66: Why, beyond the spectre of universal upgradability, would one want integrated ATA-66 when planning to use the 160/M SCSI interface? Perhaps a less expensive board w/o the ATA-66 controller would be a better choice.
3. 2x USB, serial, parallel: No complaints.
4. Integrated Ultra 160/M and U/W SCSI: Good choice if running the two HD 'standalone'. Perhaps a RAID controller on top of it to handle the data spool instead of the single 36 G drive. The controller and smaller SCSI volumes will probably be more cost-effective in the long run also.
5. 733 PIII: Intel may not be terribly committed to releasing faster chips in the slot 1 variety, now that they're also trying the FCs and PPGAs. If you're counting on a brain transplant down the road, perhaps a quad board with two empty slots would serve you.
6. (storage): A single 9G volume for the root fs and a 36+G volume for data is a fair start, but as mentioned earlier, RAID for the data spool may be more cost-effective in the long run. The DVD drive, well, I'll leave it alone.
7. Viper V770: Go with a Matrox, and a G200 at least. Better Linux support and a better card for the money.
8. 3c905: Go DEC Tulip-based instead. It performs better under Linux, and the driver is excellent.
9. SB PCI128: Try a standard SB16 instead. You'll never notice the difference.
10. Keyboard: Please, if there is to be someone typing at it night and day, substitute a more expensive ergo unit.
Remember folks, they'll be spending OUR tax dollars on this, let's give them the best bang for the buck!
.sig: Now legally binding!
This seems more like a wishlist for a home PC than a number-cruncher for data analysis. Couple of points to keep in mind when specc'ing hardware for a professional installation (corporate, organizational, or even millitary).
1) Anything you buy today will be hopelessly obsolete in 18 months: that's the PC life-span. Live with it.
2) You will -not- have the time to upgrade these machines. It seems cheaper on paper to swap out video, drives, and motherboards to and from a generic PC. It's not. Factor in: downtime, reconfiguration, getting the new stuff to work, filling out the paperwork for upgrade components, etc, etc, etc. Forklift upgrades are the -only- practical way to run a professional shop. Drive space and RAM should be the only things you touch.
This comes from long personal experience. You can afford the time and effort to screw around with a pieces-parts home machine. It's a whole other story when you are on the company (gov't.) dime.
3) Established, standard components all the way. The bleeding edge is for chumps and loosers and guys with nothing better to do than hack device drivers.
This -includes- OS! Your system vendor should install and support the version of Linux you want to run. No ands ifs or buts.
Bearing these three principles in mind, I would reccomend buying the fastest possible systems you can afford, and go with reliable, mid-range components in standard configurations. Standardization is -key-: bits'n'pieces is no way to run a high-end shop. To achieve this end, talk to VA Research about their Xeons, Penguin Computing about their Athalons, and Microway about their Alphas. Go with their reccomendations, and avoid customizing their standard configurations.
I highly reccomend Alpha hardware for speed, reliability, and compatibility. (Second only to x86) The Alpha/Linux combo has been proven in high-end applications like cinema special effects and weapons engineering supercomputers. (What do you -think- those Beowolfs are used for? Quake servers?)
The faster the system, the longer you have until it obsoletes. Another point in favor of the Alpha. The more vanilla the component, the easier it is to replace in case of failure. Standardize, standardize, standardize!
SoupIsGood Food
Even 2-D, what if you were doing 1600x1200 at 32 bpp? Would extra RAM be wasted?
I guess the real question that needs to be asked first is WHEN you intend to purchase these systems? If you are aiming at using cutting edge technology but don't intend to roll out for 6 months, these systems will be old hat by the time you get them out the door.
Things to note:
If you are doing lots of data collection, make sure that you use decent spec drives and controllers, and that where you collect your data is on a totally seperate controller or channel from the main system drive(s). I'd look at a RAID controller (64 Bit PCI?) to provide a RAID striped solution to reduce disk rotational latency (not to improve data protection) and using a few smaller, faster disks. Don't use too many disks to allow for future expansion while maintaining speed. The idea is to dump the data to the drive as fast as possible. Look at the Alphas and their 64 bit PCI buses. (This is also apparently going to appear in the Athlons, so keep an eye on that). There are also Intel's that have 64 bit PCI buses, though they are no where near as common, and I have no idea how well they are supported under Linux.
Dual/Multi-CPU Athlons are round the corner. They will have much more bang for the buck than Multi-CPU Intels, especially with FPU bound applications. Also remember that while the Intel CPU's all SHARE the FSB, the Athlons each have a dedicated path from chipset to CPU, which could avoid a very problematic bottleneck if you are doing lots of data crunching. Note: I'd also wait for the Athlon's with Integrated Cache on-chip, as this will drastically improve performance with large amounts of data.
If you're looking at using OpenGL to render data visually (which is what I assume you've been doing on the SGI's), then I'd suggest to wait for XFree 4.0 which is also "just around the corner". The TNT2 Chip from Riva is very well supported, and is definately up there in the performance stakes. It is also not swamped with problems like the G-Force (NV10). You might also look at hardware OpenGL cards, as there are a number that are supported under Linux.
If you are using sound a lot, (eg: for audio data capture) then you will want to talk to the people who wrote the drivers, or at least Alan Cox. Figure out what card provides the services you need with the least CPU cycles. Go with this, don't go with something generic simply because it's cheap. Weigh up the differences. Also beware of older SB PCI-128's, as there is a lot of stock out there that are rebadged Ensonique's (who Creative bought, and effectively designed the PCI-128) that can cause problems. I've also noticed that SB PCI-128's have a large failure rate (my work PC here has one that just makes burbling sounds instead of music).
Whatever you do, if you are aiming at cutting edge, get an idea of your time frame, and work to what will be available then. Trim the edges, but don't trim too much, particularly if you want systems that will perform exceedingly well in the future, and remain upgradable. This is the mistake I see every government department make around the world. They change something they believe isn't as important, and by changing one small part in the system, they effectively cripple the system performance.
Either way, good luck!