Slashdot Mirror

We do 3D image reconstruction - the biggest techbological advantage for us is 64-bit, but the extra memory might come in handy for volume visualization - basically when you try to create a translucent image of the voxels in a cube and be able to rotate it in real time. That can be done with textures.

If you want a lot of RAM, try the SGI Altix 3000. It's a server (not a workstation), but you can have up to 24 TB of RAM. Yes, that's 24 TERA BYTES.

Silicon Graphics sells the Prism, you can get up to 6.1TB of RAM on some models.

Gobs of memory & Linux

Gobs of memory & HP-UX

Gobs of memory & Solaris

Thousands of phamaceutical, oil and research companies around the world use this kit to get results, so why can't you?

That was the best line in the whole movie :-) "I know this, it's UNIX!"

But what they should was a "file system navigator" that is real:

3D File System Navigator for IRIX 4.0.1+

Companies like SGI spend a lot of time and money working on visualization systems that allow for multiple people to be immersed in a synthetic 3D world. SGI's Reality Center wall room systems are quite impressive.

Companies like SGI spend a lot of time and money working on visualization systems that allow for multiple people to be immersed in a synthetic 3D world. SGI's Reality Center wall room systems are quite impressive.

Companies like SGI spend a lot of time and money working on visualization systems that allow for multiple people to be immersed in a synthetic 3D world. SGI's Reality Center wall room systems are quite impressive.

Companies like SGI spend a lot of time and money working on visualization systems that allow for multiple people to be immersed in a synthetic 3D world. SGI's Reality Center wall room systems are quite impressive.

Companies like SGI spend a lot of time and money working on visualization systems that allow for multiple people to be immersed in a synthetic 3D world. SGI's Reality Center wall room systems are quite impressive.

You'll have a hard time convincing anyone in-the-know to run Linux on anything that isn't x86.

That is complete crap. I'll give you a couple of people who use Linux on non ia32. SGI, IBM, plenty of ARM products, Alphas used to be (not sure if they are still sold).

So I assume you'll say you know better than all these people because you have some ancient SGI MIPS machine that nobody cares about and Linux doesn't work on it, right?

How about you show a few examples of people in-the-know who run NetBSD. Yourself doesn't count.

Its strategy towards porting is "convince the rest of the kernel it's an i386 and work like that", which fails on a lot of systems which are fundamentally different from i386, even if they have some things in common (ISA isn't the same everywhere, for instance).

Sorry, Linux is ported to more CPU architectures than NetBSD, including architectures without MMUs and PPC64 (which doesn't have pagetables), neither of which NetBSD can handle.

So give me a single example of something that is i386 specific (or even i386 centric for that matter) in generic kernel code, and I'll eat my hat. If not, you are just a clueless troll.

The same clean code and clean design that allow this kind of abstraction lead to a generally clean and stable system. NetBSD's worst stability problems occur only in device quirks which haven't yet been fully understood (you'll notice Linux has the same quirks but the hacks around them are usually done earlier, since Linux contributors don't care if something is a hack or not).

I don't think you are in a position to say that Linux kernel developers don't care if something is a hack or not. You're nowhere near in their league. But give me some examples of these so called "hacks" (clue: a workaround isn't a hack by definition).

Where the hardware is non-quirky, the system simply does not encounter problems. Simple as that. You could say the same for Linux to some extent, but for Linux, achieving stability is all about quirks for everything, even where it's not needed [see the first mention of i386 'emulation' above].

I'm sorry, but you are simply wrong about this.

That's why the Linux kernel is an order of magnitude larger than NetBSD's but does not have the functionality to justify it.

The Linux kernel has an order of magnitude more device drivers, supports more CPU architectures, has orders of magnitude more filesystems (including proper journalling filesystems), can be configured to run in 2MB of RAM, or scale up to 512 CPUs and terrabytes of RAM, thousands of disks, hundreds of PCI busses...

True, you can build very large clusters from these bricks, but the bricks themselves don't scale beyond a relatively small number of CPUs.

NASA's Columbia cluster ^ 512-way SGI machines running Linux (actually 20 of them...) Not to mention "Columbia's record results were achieved running the LINPACK benchmark on 8,192 of the NASA supercomputer's 10,240 processors. Columbia also achieved an 88 percent efficiency rating on the LINPACK benchmark, the highest efficiency rating ever attained in a LINPACK test on large systems." from http://www.sgi.com/company_info/newsroom/press_rel eases/2004/october/worlds_fastest.html

A Porsche

I know what you mean. When we worked on an IMAX film we had to break the images up into 4096x4096 parts in order to get the larger 19Kx19K sized images. We too, were limited by the OS's ability to ramp up past 2GBs in both memory as well as disk drive files.

The thing is - we now can do those huge films with ease because of the work done at IBM as well as at SGI. Their filing enhancements (and ext3) allow us to have huge files. The memory problem though - still exists.

As others have pointed out - Windows XP should handle these problems (although the disk drive problem of not really removing the directory entries will bite you sooner or later [and for evreyone out there I will try to find the SGI white paper on the problems of Windows XP's disk drive NTFS methodology]).

Anyway, gotta run!

[This is my last reply (not anonymous and karma-free) to this name-calling AC]

That one by Oppermann was a mailing list post, in which he was implicitly referring to a Xeon. Not surprisingly, the submitter of the /. story understood that. In fact he says:
"Andre claims that FreeBSD can now route 1Mpps on a 2.8GHz Xeon whilst Linux can't do much more than 100kpps."

Notice the "on a 2.8GHz Xeon" part... *Both* the numbers (1 Mpps and "not much more than 100kpps") are referred to one particular architecture.

And, I repeat, there's *nothing* in the post you linked that either disproves that, or justifies your previous claims. The 1.3 Mpps figure is *not* referred to the same architecture. About the Xeon, the post you linked just says "Our numbers on Xeons are less than 1Mpps".

I think that's enough to acknowledge that if there is a troll, that would definitely be you.

One more thing about the "can't do much more than 100kpps" part: even if it were an underestimate, and the actual figure turned out to be 300, 400 or even 500 kpps, it's still hardly relevant compared to FreeBSD's 1 Mpps.

Not to say that FreeBSD is without weak points, of course (just like every OS, and the devs are surely working on those), but it routes much better than Linux.
And underlining this fact, in the one place (/.) where trolls have been spreading the "BSD is dying" FUD for *years*, is more than appropriate. :)
--
Being able to read *other people's* source code is a nice thing, not a 'fundamental freedom'.

here is a simple post showing Linux can do 1.3 Mbps on a dual opteron of similar capacity. Even if Linux is not as fast as FreeBSD in routing, it shows that Oppermann is talking out his ass when it comes to Linux performance. Where did that 100kpps figure come from? Absolutely no basis for the claim whatsoever.

You would have believed this was real? That 3D Gui from Jurassic Park runs on IRIX

What about the Prism http://www.sgi.com/products/visualization/prism/?
Did anybody test one of these?
Might the Raedon-driver profit from SGI?

It was FSN, and the machine was an SGI box, not a Quadra (even if it were, an X server could conceivably been used).

Futhermore, even though the scene was laughable it was not that far fetched. Remember she didn't say "this is a UNIX system because of the 3D file manager," presumably she identified it by the file system layout.

I mean really what does UNIX look like? What one thing has it ever looked like? OpenWindows, X, CDE, different shells. at all levels except the POSIX API and a bag of utilities and a somewhat standard file system layout, there is nothing visually distinguishing about "UNIX" (even if you restrict UNIX to a particular implementation like Solaris or IRIX).
The test of a true Unix weenie is that he/she is able to operate the system through all the customized/weird interface variations. I guess everyone is just intimidated by the JP chick.

It's called FSN, and it's right here

http://www.sgi.com/fun/freeware/3d_navigator.html

It's not common but it existed.

the 3d interface from jurassic park actually exists and you can d/l it if you want, it never was completely finished though and is thoroughly dated by this point.
you can find it here

Ulrik

You can download Irix 6.5.24 (I think) for free from the SupportFolio site (http://support.sgi.com). Just create an account and you'll have access to these updates. I have two O2 myself (an R12000 and an R5000) and the Irix 6.5 CDs, so I used the updates on that site to get to 6.5.24.

Why not just download them from http://support.sgi.com? Supportfolio accounts are free and provide access to OS updates. The latest version is...(checks account)...6.5.26. Since you already have the 6.5 CDs you can just install 6.5.0 and then using inst or swmgr to upgrade to 6.5.26. The harest problem I've run into is running out of drive space during the upgrade (SGI likes to stick tiny OS disks on their machines--especially those old ones).

inst (and its X frontend swmgr) are among the best software installation managers I've ever used. swmgr is pretty intuitive. It's certainly a whole lot easier to use than RPM (try asking an rpm newbie how to find what package installed what file, or where a package is going to put its files for instance).

The monitor's specs are a bit suspicous - the article claims more realistic 6400x768, which is 5 screens with 1280x768 side by side, the higher resolution is probably just downscaled. And it's just that, 5 monitors, side by side, with the borders between them. With 6400 pixels wide, does that even work with opengl or directx over multiple graphics cards in 3d? It's quite some time since I checked the last time, but back then, 3d applications over multiple graphics pipes weren't really a standard feature, and certainly not available on PCs.

I say either get one of those sgi / barco reality center 3300W things, which have a more useable aspect ratio, higher resolution and, if you ask nicely, come in a faster refresh version for use with shutter glasses, or build a cave. Damnit, if it's going to be obscenely expensive, then at least do it properly instead of jumping for some shiny new toy.

You are thinking of pointblank. We played this every day at lunch hour. God how I loved that game.

You do realise that that the Jurassic Park thing is a real SGI/Irix file manager called fsn?
The system it models is something different... but it is a quite real Unix filemanager.
Though I do agree that many people attribute some kind of magical qualities to computers that don't exist. As Arthur C. Clarke said "Any sufficiently advanced technology is indistinguishable from magic."
I'd also like to offer a corrolary:
"Any person insufficently advanced is unable to distinguish technology from magic." I think that this is the case with most of the general public that watches crap like CSI and believes all of it.

The application she used was a real sgi application called fsn. No, it's not generic unix, but SGI doesn't get to sell multi-thousand dollar graphics cards if all it supports is "generic" unix.

I think of Jurassic Park where the little girl is staring at the computer with some 3D file system view and says, "This is a UNIX system, I know this" and I realize that most shows are not very accurate. I imagine CSI's view of forensics is about as accurate as Jurassic Park's snapshot of UNIX. But it is entertaining anyway.

That however was indeed a Unix system, running SGI's 3d File viewer called FSN

By "Power970" you seem to be referring to the PowerPC family. Don't forget that this chip family is based on the Power architecture from IBM (with some help from Apple and Motorola). The Power architecture contains other chips too, some of which don't have the limitations you cite. Certainly the chip architecture is fully capable of supporting machines with a larger number of CPUs running a single system image.

Although the really big (and custom) Blue Gene systems are apparently clusters, there isn't anything about the IBM Power Architecture itself that would prevent large monolithic systems from being designed and built.

The SPARC architecture can be used for machines like this, too. (Remember the CM-5?).

Building a supercomputer with a large number of CPUs running a single system image is a unique task with a limited client base, and SGI has experience with that. A whole lot more than CPU choice goes into making it work. The way they tell it it was quite a rush. The internal conversation must have gone something like this: "OK, team, we're going to build exactly one of these, and we already decided the price!" NASA doesn't build rockets like that, but SGI can build supercomputers like that. Impressive.

SGI deserves kudos. But if we step back and look at the big picture from the vantage point of SGI, it sure looks like SGI chose the IA-64 CPU for marketing reasons, not technical reasons. I'd have to guess that their engineering tasks would have been made easier by using a CPU that draws less power, for example. They've been on the ropes for years and conventional wisdom says to back Intel if you're in trouble because that's the safe bet for marketing. Why this remains conventional wisdom when the track record clearly shows that UNIX vendors who switch to Intel are cut up and fed to other UNIX vendors, is another topic.

You're right of course, that there are two different classes of super computers on the Top 500 list, with one class based on the cluster concept, and the other based on the concept of a single system image. Clusters are radically less expensive, and monoliths are better at certain computing tasks, and it's hard to compare them.

Monoliths often get custom case mods, though, and thus tend to look cooler. Who would hang a poster of a beowulf cluster of generic beige 1U rackmounts on their office wall? Everybody wants a poster of a Cray or a CM-5 or a Mach 5...

Hey! I just realized monoliths don't seem to look as cool as clusters lately. What's up with that?!

aahha, like 'fam (file alteration monitor)' we have in *nix :-)

.. anyway, to my recollection (I'm not a Mac-guru), but the FS they use, HFS+, stores 'Resource forks' and 'Metadata' strait into/in the FS itself (granted, exactly what sort of type this 'Metadata' is, I don't know...might be other sort/type of 'Metadata' then what we are discussing here...)

If we're talking solid state disks (in which direction in my opinion this is pointing to), I'd rather see something like this in my household :) As to the size of such a "vibrating" storage solution... well, if I don't see it, I don't mind, but I hope it won't cost too much, it won't need more power, it will have higher lifespan, and at least two of these seems highly unlikely (just pick :)

All in all, just let them boil a bit, let's see what comes out. Yup, one more thing, hopefully one will be able to cary home a >100gb version of such a thing in one's hands :)

Bullshit. The sgi machine was up at NASA doing real work. The IBM machine's benchmarks have only been repeated in their manufacturing floor. I'm not saying the results are bogus but I'm saying that sgi is the one with the production ready supercomputer.

The bigger deal is that this sucker was put together and running customer code in 15 weeks. The build went so smoothly that people were surprised when a freshly installed machine didn't just run by the end of the day.

No, it just means that IBM (this time, at least) are full of PR-shit. The supercomputer they're talking about has one big difference from NASA's new system - the IBM box hasn't even left development yet, whereas the SGI system has been shipped and installed at a paying customer's site.

Need proof? Here's one way you could go about getting it:

(you) Hi IBM, I'm thinking of buying a BlueGene system for my lab, but I'm wondering - what operating system can I install on it?
(IBM) It runs Linux!
(you) But I looked carefully through the Linux source code, and couldn't find any mention of BlueGene systems. Are you sure it'll work?
(IBM) Actually, you need to use our modified version of Linux.
(you) Oh okay, no problem. Where can I get the source?

At this point, the IBM sales rep's head will implode as he realises he has only two options: he can either stall (and the longer he stalls, the more it looks like he might be violating the GPL!), or he can admit the truth: the operating system doesn't _actually_ exist yet in any sort of finished form yet.

On the other hand, SGI's version of Linux that runs on the 256-way Altix BX2 systems is a free download, as it should be. Start here.

Hope this helps! Your post does raise the interesting question of where the Top500 group (or benchmarking people more generally) should drawn the line: it's nice to know about cutting edge machines, which may even be "one of a kind" installations, but somehow it's not so helpful when companies announce performance results of things they haven't even finished building it. It would not be out of the realms of possibility that the final BlueGene machine shares little (or nothing) in common with the machine they're talking about today, which would mean that this PR is basically nothing but a fairly baseless grab at mindshare.

Alpha based workstation with OpenVMS or Tru64

Sparc based one with Solaris9

Dual G5 Mac with OSX. say no more.

SGI ws with IRIX

HP (parisc)with HPUX

and last but not least an x86 compatible possibility:

Orion DS-96 Deskside Cluster Workstation. Yes, thats the number of CPUs in it.

FYKI

ps: no, I cannot afford them either. Yes, you could run Linux/BSD on all of them.

Unix to Linux migration makes sense.
Just remember: not all Linux platforms are created equal. Just consider:
Moving from Unix to the world of Linux gives you new control over your compute environment, at a lower cost. However, not all Linux platforms are created equal.

What are the growth/scalability limits for your platform?
Can you increase only the compute resources that you need?
Will you have to call multiple vendors for support?

The SGI Altix family of servers lets you run Linux with no limits. Unlike other systems, Altix scales from 2 processors up to 256 processors in a single Linux system image, and enables you to independently increase I/O, memory, and processors for perfectly right-sized systems. Altix is also easy on your budget, with the ability to start small and grow big, all in a unified architecture - no more "forklift upgrades"! With over 20 years of industry leadership in developing high performance systems, we know what it takes. We've helped to extend the Linux kernel with performance optimizations. We also know that you want to go to a single vendor if you have a system problem, so will be your single point of contact for system problems - regardless of the source.

So, expect the best - SGI Altix.

SGI just released their new visualisation workstations named Prism. In short, its an Itanium with ATIs running Linux.

Wow, I hope they catch the SOB who stole all the RAM

--
I've got my iPod now, and my PC is shipping right now. Thanks!

One idea was to link thousands of dual-processor commodity servers into a sprawling cluster, but NASA quickly dismissed that approach. "We're trying to solve some of the toughest scientific problems in the world," says Jim Taft, task lead for the NAS Division's Terascale Applications Group. "We needed a system designed to efficiently execute the algorithms used in NASA's premier science codes, rather than one that would merely do well on artificial benchmarks."

Brooks and his team instead pointed to Kalpana, an Intel® Itanium® 2-based, 512-processor SGI® Altix® 3000 system in use at NASA Ames since November 2003 and named to honor Kalpana Chalawa, a NASA scientist lost in the Columbia accident.. In less than six months, Taft says, the Kalpana system - the first 512-processor Linux® system ever to operate under a single Linux kernel - had revolutionized the rate of scientific discovery at NASA for a number of disciplines.

On this picture you can see what I'm sure is an 'Intel Inside' sticker on the bottom of some of the cabinets.

Download it here.

Check out http://www.sgi.com/products/storage/ for some more info about the storage they are using. For those that don't want to wander around the site, there is a link under the picture of the storage array that says "Watch a Video" and it gives an overview of the technology that SGI uses in their storage solution.

They use tape storage from Storage Tek like this one
And harddrive storage from Engenio (formally LSI Logic Storage Systems) like this.

Wow, I'm glad I'm not the one installing 10240 heatsinks..

Here are some much larger, high-quality images.

http://www.sgi.com/company_info/newsroom/virt_pres skit/project_columbia/columbia_images.html

And with all that power they feel the need to .ZIP their .JPG images which actually shaved off an entire 4K on this single 848K file. Wow. I should have thought of that.

Columbia

This page contains images of the NASA Altix system. After reading the article I was curious as to how much room 10K or so processors take up.

1) This was fully deployed in only 15 weeks.
(Link)

2) This number was using only 16 of the 20 systems, so a full benchmark should be larger too.
(link)

3) The storage attached holds 44 LoC's (link)