Slashdot Mirror

intel and SGI gave a 10240 itanic cluster to NASA.

If that's so, why does the machine make a notable appearance in SGI's 2005Q2 financial statement?

Allow me to humbly suggest you either call them on fraud, or admit you're full of shit. AMD astroturfer, are we?

> Content-wise, there wasn't really much there for me.

Yeah, I doubt there was anything in it for anyone interested in filesystems.
And seeing XFS is my day job, the mistakes were pretty obvious, too.

One, a b+tree does not make a filesystem.

Two, in all that talk about b+trees in XFS, he made some basic mistakes. There's
only one inode b+tree per AG, there's two extent free list b+trees per AG, and
the superblock has no b+trees in it at all. And they are used in many other
places in XFS as well.

Three, there is so much in XFS that this didn't even mention that it could
hardly even be called an overview of XFS, let alone design. I mean, the overview
page for the XFS project (http://oss.sgi.com/projects/xfs/) has a more informative
description of XFS than this article....

He said "main servers" not all server. Linux is flawed because it is a young OS. When you have something that you run for two decades, you don't want to switch it just for the sake of the switch. That is dangerous.

Anyway, the 10240 processor system running Linux is then not a "main server" according to that person,

http://www.sgi.com/company_info/newsroom/press_rel eases/2004/october/worlds_fastest.html

XFS does things that ext? and Reiser can't do. Reiser does things other FSes don't do as well. It's a true 64-bit filesystem and it supports insanely large filesystems, up to 9 million terabytes in 64 bit mode (with a 64 bit kernel.) It even provides realtime support, although I guess that's still beta in linux? It can be defragged and even dumped while live. It has insanely quick crash recovery. And of course, it does other stuff too; check the project page. XFS may not be the fastest filesystem - it may even be the slowest - but it's got features no other filesystem has. If you need them, XFS is the winner. Hell, if you just trust XFS more than you trust other filesystems, it's the winner. (Sorry, but I wasn't sleeping when reiser was eating everyone's data, and ext3 handles corruption much more poorly than any of the other Journaled options.)

Some of us like having zero-overhead generic container classes. It comes in rather handy in practice.

There is no such thing as a perfect language.

Welcome to IRIX circa 1995... remember fsn?

Okay here you go It's from IRIX
Don't ask why I know but I do.

just realized the girl in jurassic park predicted the future by saying those lines...she was looking at a Mac I believe and said that line....I could be wrong though....speilberg knew...

She was looking at an Irix OS with a 3D file navigator.

Looks like an idea that been around the block for a while.

1994

Paul Haeberli did this in 1994 - it's a neat idea for some experiments, but not a very practical solution.

To give an idea, 128 nodes will give you at most 512 processors (more becomes EXTREMELY inefficient). 512 processors will net you a place between 300 and 500 on the current top 500 list. This will be very different on the list to be released six months for now... such small clusters might not even show up.

Then there is the user group of HPC systems. It is a VERYsmall market, with a userbase, a group of anministrators and a group of manufacturers traditionally used to UNIX, and now migrating to Linux in droves. Windows is not even duscussed. The announcement of the Windows Compute Cluster edition was cause for great hilarity at the workplace, where jokes like parallel word/excel and high-performance visual basic started floating around. No one will take Microsoft serious in that market.

Perhaps Microsoft will sell some systems to some manufacturers, like in the automotive or pharmaceutical industry. But these guys already know the ways to traditional vendors selling them Linux clusters, vendors like SGI for instance. CHeck the SGI Manufacturing page.

So... will Microsoft compete? So far, they announced an operating system for clusters. Important questions remain:

• Will the OS run headless?
• What low latency networks will be supported?
• Are MPI and OpenMP implementations available?
• what about remote management, remote login and remote copy? (On a side note: why is it that Windows 2003 can't have simple stuff like ssh and scp built in?)
• what applications will be available?

...by 80 Gb/s Craylink cables. I think this interlink technology is also licensed to SGI for use in their Origin computers.

NUMALink

Have you got any links to some page describing some configuration similar perhaps to your colleague's system? I haven't heard about 10GiB/s CrayLink and find it very intriguing.

sgi's itanium based system has so far supported 10240 processors. They've already left irix in the dust.

SGI made some great machines both in the form of the hardware and the looks of the hardware. They also provided us with the likes of OpenGL.

The problem is that the market they once had, being high-end graphics workstations, is being eaten up by cheap MS-Windows based systems. They could try redefining themselves, but I not sure what form it could take. While their version of Unix had some nice additions, it was never really a selling point. Their cheapest systems start off at $9000, which more expensive than Apple, and they also have less technology diversity than a company like IBM to help buffer any slow growth of their hardware. Maybe if they offered a very capable $4000 machine, it might help them attract people who might have never considered them before?

BTW CATIA, which is a very important piece of CAD-CAM software in the automotive and aeronautical industry is actually Windows centric, so they benefits of a SGI machine there is zero.

They made the world a better place though, IMHO.

They made the OSS world a better place, at least. SGI is putting lots of resources in OSS software. They gave us things like XFS. Their engineers are part of the group of programmers who made (and are still making right now with patches being merged in each release) possible to make linux scalable in big SMP boxes (ie: their 512-CPU boxes). They gave us things like GLX (the opengl xservers glue)

Linux users owe SGI a lot. They're still not dead though, I hope they find a way to make SGI profitable again...

They made the world a better place though, IMHO.

They made the OSS world a better place, at least. SGI is putting lots of resources in OSS software. They gave us things like XFS. Their engineers are part of the group of programmers who made (and are still making right now with patches being merged in each release) possible to make linux scalable in big SMP boxes (ie: their 512-CPU boxes). They gave us things like GLX (the opengl xservers glue)

Linux users owe SGI a lot. They're still not dead though, I hope they find a way to make SGI profitable again...

SGI put out some increadibly cool technologies:

OpenGL - a very important 3D API

The Standard Template Library

VRML which gave rise to X3D Open Inventor which is a C++ wrapper around OpenGL.

Pretty purple boxen that were great in their day.

It seems that these came out years before the average user could really leverage them - years before anyone (including SGI it seems) knew what to do with them.

It seems a shame that such a brilliant company could have such a hard time making money. They made the world a better place though, IMHO.

These things already exist. You can buy PCI plug-in modules from Nallatech (http://www.nallatech.com/). Complete systems are also available from SGI (http://www.sgi.com/company_info/newsroom/press_re leases/2005/september/rasc.html).

From what I understand, it is still very hard to program these things, but software techonology is starting to catch up. SGI has a version of GDB which is FPGA aware.

Sure, it quite possibly *can* be done. In the same way that you could theoretically build a spaceshuttle from a T-Ford and lots of old spraycans!
When you're aiming for extreme solutions, you have to use quite extreme components. Think http://www.sgi.com/products/storage/ http://h18006.www1.hp.com/products/storageworks/xp 12000/index.html or even http://www.sun.com/storage/highend/9990/index.xml.

The only imaginable way to get petabyte on commodity hardware that I can think of is to build a seriously huge Beowulf cluster. But putting a singular FS on about ~500 seperate computers in a cluster is rather madness...
Take the good advice many fellow slashdotters has made, do NOT use a singular filesystem that spans 1 PetaByte or more...

The filesystems going to be the hardest component of this. I know of no open-source fs that could handle this. I'm assuming this is all online storage, and there is no desire to nearline it to tape. Ideally, you'd want something that could contcatenate multiple LUNs (of RAIDed storage) without having to run through a volume manager. Nothing agaist volume managers, but it'd be another component to support. Looking at proprietary FSs, you've got CXFS from SGI, which could easily handle the PB requirement and plays nice on Linux. Sun's got QFS, which would max out at 1PB and could do the volume management bit easily. Linux support was a little flakey last time I used it, but it's a free download and evaluation, you could go get it right now.

IBM's SAN-FS would also meet the capacity needs and would have the advantage of providing nearline capability, if you're into that. Sun's SAM-FS is basically the QFS product with nearline-to-tape capability. Linux is only supported as a client OS there. Of course, if you buy the mantra that Solaris is 'open-source,' then that might not be an issue.

As for hardware with any of the above solutions, you're going to be looking at using multiple RAIDing disk enclosures of some kind. At a budget, probably SATA disks talking to the controller, and iSCSI to the host. FibreChannel to the host would be a little more costly, but might be worth it since iSCSI is just getting mature enough to be usable in production.

Actually, *real* games, such as commercial flight simulators and planetariums, do benefit nicely from multiple processors. Many will use libraries that can do this almost automatically - SGI's OpenGL Performer for example.

Even some big entertainment games, like DisneyQuest, use OpenGL Performer to take advantage of the many CPUs.

SGI had state threads library since long http://oss.sgi.com/state-threads

You're absolutely right, google's load isn't that demanding on the operating system.

Ask these guys what it takes to get a cluter of 20 512 CPU supercomputers with 4TB RAM to run stably under load.

Or I guess you'll tell me NASA doesn't mind rebooting once a week, and having their systems go down in the middle of a month long simulation?

And no, they don't use custom software, they use SUSE SLES9 for the kernel.

SGI has a neat solution to this problem called DMF. It will automatically migrate infrequently-accessed files onto tape, and 'recover' them when accessed (takes a little longer than normal) automatically when there's a tape library, and I expect it will prompt an operator to load a tape if the tape isn't online.

I've no idea what the cost is, or if there's a low-end solution.

Make that four.

SGI are still in the UNIX business.

Make that four.

SGI are still in the UNIX business.

Hasn't similar technology been available for years? Like this:
http://www.sgi.com/products/appsdirectory.dir/irix /products/c/957266.html

Your criticism of Java for making early, hasty changes is dead on, but I would like to think that the damage was lessened because the changes were mostly confined to libraries - deprecating APIs and the like. Had the changes been to language features, things would have gotten much nastier (as if AWT->Swing wasn't nasty enough!)

But all of these proposed changes are to the C# language proper, not libraries. I think that's an essential difference.

Stroustrup (did you call him "its inventor" because you couldn't spell it either? I had to look it up!) wrote in that link "Library extensions will be preferred to language extensions." Hooray! IMO, it's too little, too late for C++, but I'm happy that he's made that distinction. I hope that the C# folks get that message soon, because they haven't yet.

Compare their approach to LINQ (from, where, select, in, etc. all become keywords) to, say, Apple's new Predicates stuff, which accomplishes the same thing (or at least appears to) without relying on new keywords.

Regarding C++ sublanguages - ok, my Mozilla link wasn't very good. But I stand by my point, that any given project will cherry-pick C++ features to construct their own internal C++ subset. Boost makes heavy use of exceptions and templates; wxWidgets doesn't. The STL barely uses inheritance. If you're programming for Boost, you should try to avoid using catch to catch exceptions. If you're programming for SGI, you shouldn't use C++ strings. Etc.

Java doesn't have issues like these. Java programs by and large all make heavy use of exceptions and inheritance, and recommending that you avoid catch() or strings is unthinkable! Keeping the language features trim and widely relevant is a big help.

You're right that, in principle, new users can learn best practices and start writing in the corresponding shared subset of the language, and really good C++ programmers do. If it weren't so damned hard to become a really good C++ programmer, this might actually happen in practice. Until then, best-practice libraries like Boost will remain out of reach for the majority of C++ programmers

(Though Stroustrup acknowledges that C++ makes things "unnecessarily difficult" for newbies. Maybe there is yet hope.)

There are 2 other significant players in the FS field that Hans doesn't mention:
XFS (from SGI) and GPFS from IBM.

GPFS has a different focus, but XFS seems to be aimed at solving similar problems as ReiserFS (scalability, high performance, journaling).

Well, are SGI workstations available for low cost and have OS's geared toward the average user?

No. But the real world is what businesses use, not what someone uses at home.

You can't really compare modern Linux distributions running on cheap commodity x86 hardware to SGI workstations.

Oh? ;)

People should have been using FBSD for wifi access points and aggregation long ago. Glad to see the corporate folks have finally figured out how linux errodes the bottom line.

The
stupider
ones
haven't
figured
it
out
yet.
You'd
better
go
tell
them!

You have to realize, that if Apple does nothing, Microsoft and Linux are going to be battling it out over the next generation of desktops running on ultracheap commodity hardware. Apple has a unique opportunity here to jump in and really make some inroads into that marketplace, perhaps even to become the major player. They have a mature, solid and easy to use product that would be way ahead of either Microsoft or Linux in their own marketplace, if it could run on that same commodity hardware.

Apple could do this, but in my opinion for commercial reasons they should not and I'm prepared to bet they will not. Apple is a premium brand. Their ability to make a profit depends on it being a premium and an aspirational brand. That brand is built on images of style, quality, ease of use, reliability. It's the quality feel which is Apple's USP, Apple's raison d'etre.

Apple cannot make a profit as a software vendor head to head with increasingly good Linux offerings and the market dominance of Microsoft. Apple have to make a profit on their hardware. But it's their software which, for the average user, actually sets their hardware apart, which gives it its face. If Apple customers can get Apple software on generic hardware, that's going to hurt Apple in two ways: they'll lose the premium hardware revenue stream, which is important to them, and their users will experience unreliability and glitches which are more-or-less inevitable in a commodity hardware environment where many of the device drivers are being written by half trained monkeys over whom Apple have no quality control sanction. And it's Apple's brand image which will be hurt by that.

In my opinion, Apple should look carefully at the fate of Silicon Graphics. SGI, for those who don't remember, used to be a niche maker of UNIX boxes mainly for the design, visualisation, and media markets. They made very very nice hardware, very high quality, running their own UNIX and based on their own MIPS family of RISC processors. They were, as Apple is now, an aspirational brand - a maker of premium and very desirable machines. Then, in the late nineties, they lost confidence in what they were doing and started to ship intel boxes with Windows NT. It was disasterous for them. They couldn't maintain their premium and they diluted their brand value. They've been forced back into their core business - high end visualisation systems on MIPS hardware - but as a much smaller and weaker company.

I think Apple are taking a very big risk with their brand image by moving to Intel at all, not because there's anything wrong with Intel processors as such but because it dilutes their brand differentiation. I think it's vital to them to clearly differentiate between an Apple machine and a generic PC. Even if there were no technical reasons for not using generic hardware, I think there are the strongest possible commercial reasons.

Really? I guess it depends on what you mean by compete. SGI's Altix runs quite nicely at 512 CPU's per node. Significantly more than any Sun system from the point of performance and scalability. Solaris certainly has other advantages, etc, but my point is to qualify your statement of "can't compete", because in some cases you're right, the competition is over and Sun never showed up. Each architecture has it's advantages.

It is true that Sun seems to have made some improvements. Weren't they supposed to fix ufs with Solaris 10? DOn't get me wrong. Solaris pays a lot of bills.

Solaris scale better then the others and runs well on 64-way multiprocessor systems

This Linux box scales to 64 processors. So does This AIX/Linux box.

http://www.sgi.com/products/servers/altix/
"The SGI Altix 3700 Bx2 supercomputer supports globally addressable memory across multiple nodes, scaling to hundreds of Intel Itanium 2 microprocessors. Each Altix 3700 node can combine up to 512 processors in a single Linux operating system image, and is ideal for the most complex HPC problems, enabling innovation without limits for technical users."

Say what you want about the OS, it CAN be secured and has given me the least trouble out of any vendor.

If you don't like the OS, they have a line of linux servers too. Pretty nice hardware.

One common misconception is that these are 64 way SMP boxes. They aren't. See, they're really 4 CPU bricks (C-bricks) with low latency, high bandwidth interconnects and smart memory management units to do memory references on remote boxes.

You most probably know better than I do (as I have never worked on an Altix), but SGI seem to disagree with you :

Up to 512 processors tightly coupled and operating under a single copy of the O/S

While Linux's SMP is perfectly good, I sincerely doubt it would scale quite as well as the boilerplate enterprise OSs such as Solaris or the mainframe platforms.

Well...

.haeger

Currently, Itanium is pretty much exclusively used in really high-end machinery such as SGI supercomputers, which are not really within the reach of the ordinary computer enthusiast. So, since there are more urban legends about the chip than actual knowledge, it's easy for such myths to spread.
But there are some trends that seem to push this chip outside the exclusive markets - e.g. the aforementioned SGI is working hard to translate its knowledge in the field of supercomputers into things more accessible, such as high-performance enterprise servers. If Intel makes concessions to the price, those high-performance servers might actually use Itanium.
Of course, if Intel keeps the price high, then those "small supercomputers" will just use some other CPU, to keep the total price down.

Bottom line: it's a perception thing. It's easier to ridicule what is not known, hence the "Itanic" myth.

Are these cards compatible with SGI Prism systems? The current SGI Prism systems appear to include a ATI FireGL card.

http://www.sgi.com/products/visualization/prism/

But SGI doesn't sell IRIX machines any more. Maybe that's why they're dying.

Here now, what's this, then? And this? Certainly, I don't have a use for an IRIX server/supercomputer, but they do indeed appear to be available.

But SGI doesn't sell IRIX machines any more. Maybe that's why they're dying.

Here now, what's this, then? And this? Certainly, I don't have a use for an IRIX server/supercomputer, but they do indeed appear to be available.

In fact, I'm not sure IBM cares about beating Windows on the desktop since they sold their PC manufacturing business.

And SGI? Well, they've been on supercomputers, but really don't seem to have a story when it comes to selling high-end UNIX servers.

Why does it matter that you are a D3D programmer? Neither of your arguments have anything to do with programming in D3D, rather they have to do with the history of D3D.

Which is irrelevant. The fact that MS bought D3D is irrelevant to whether or not D3D now is innovative. Innovation is not the same thing as invention (or patentability). Prior art does not negate innovation. The fact that someone did something before you doesn't mean that you cannot be innovative. Google hasn't done anything that hasn't been done before, they just do it better. Execution matters. And the execution of D3D continues to get better and better. Meanwhile, OGL stagnates. The extensions are not part of OGL until they are approved.

As far as never doing anything first... You only think that because you don't realize how extensions in OGL are added. Do you think that someone dreams up an extension first? Not a chance. Look at the extensions registry. The majority of proposed extensions (and pretty much all of the interesting ones) come from NVIDIA, 3DFX (still haven't been approved, huh?), ATI and the older 3D HW manufacturers.

A lot of the extensions that are proposed to OGL are a direct result of requirements for new versions of D3D.

But as a few examples of things that D3D did first... Multitexturing. Shader support. Caps bits. MRTs.

That's not to say the D3D is always the latest and greatest. For example, percent-closer-filtering (used in HW soft-shadow mapping) has had an extension available in OGL for what now, 4 years? And it still hasn't made it into D3D.

The bottom line is that MS is still innovative in that the things that are available in D3D, are available on all 3D accelerated hardware. (Which is actually why PCF is not included in D3D. ATI has yet to add support for it.)

SGI has a proud history of innovation in graphics, microprocessors, operating systems, etc, but this post has to do with one other small part of that history... their cases.

Well before the iMac, SGI always had instantly recognisable hardware. I wish there were PC case manufacturers with the same vision, who would churn out something stylish and interesting that doesn't look like an Air Jordan.

My favourites: the Octane http://www.sgi.com/products/remarketed/octane/, and Tezro http://www.sgi.com/products/workstations/tezro/.

SGI has a proud history of innovation in graphics, microprocessors, operating systems, etc, but this post has to do with one other small part of that history... their cases.

Well before the iMac, SGI always had instantly recognisable hardware. I wish there were PC case manufacturers with the same vision, who would churn out something stylish and interesting that doesn't look like an Air Jordan.

My favourites: the Octane http://www.sgi.com/products/remarketed/octane/, and Tezro http://www.sgi.com/products/workstations/tezro/.

Yes, but does your low end PC allow you to:

* Unlock the secrets of the planet by intuitively grasping the complex interplay of oceans, sunlight and atmospheric effects
* Diagnose life-threatening medical conditions in unprecedented detail
* Achieve six-sigma quality by enabling domain experts to work collaboratively, not sequentially
* Extract currently unrecoverable petroleum assets through better understanding and management of existing fields

I don't think so!

(source)

Or just download a source archive and a documentation archive from http://www.sgi.com/tech/stl/download.html

Linux-on-Opteron plus the OSS tool makers are getting there, but not yet.

Linux-on-Itanium is already there. SGI would be more than happy to sell you a big SMP system with an operating system that has the features that big organizations need. Check it out.