They have, almost 5 years ago, actually. You are simply uninformed, haven't been paying attention:

http://www-03.ibm.com/press/us/en/pressrelease/19198.wss
http://www-03.ibm.com/systems/bladecenter/hardware/servers/qs22/index.html

And the IBM Cell blade has a Cell chip, the PowerXCell 8i, with 1 extra SPE, 5 times the double precision floating point performance, and with the Infiniband HCA, over 50 times faster network communications.

Where have you been these past 5 years? :-)

RoadRunner, http://www.lanl.gov/discover/roadrunner_fastest_computer

has over 13,000 PowerXCell 8i chips which are 4 times faster than the PS3 Cell chip on FP code. RR executes Linpack at 1.04 PetaFLOPS/s, just over double what AFRL is claiming for their little bullshit 1760 Cell cluster. If AFRL is quoting PEAK hardware performance, their 1760 PS3s would hit a mere 180 TFLOPS/s, far short of that 500 figure. The head nodes they mention would add another 10 TFLOPS/s peak. They didn't specify the number and type of GPUs in the cluster. Even so, they're still not going to hit 500 TFLOPS/s, nowhere close to it running any application, and not close to it if quoting PEAK hardware numbers.

Those Air Force boys must be smoking some good weed these days to hallucinate that 500 TFLOPS/s figure.

What happens when the kid has already turned off Google Safe Search, then searches for "rooster" as they discussed chickens and roosters in class that day, at age 5. Rooster gives "cock" as a synonym, so he searches "cock". Then he hits the images button. Some of what he sees _will_ be traumatizing and will prompt questions to the parents. This is a far cry from coming across an issue of Playboy and seeing boobs as you and I did as youngsters. Google and the web are a game changer here and the potential for trauma is high, as well as other consequences.

I thought the big deal with the LHC is that it was supposed to give us the Higgs Boson, as no other collider on earth was powerful enough to create the Higgs. Enough about this QGP junk. Where's our Higgs Boson?

If your SAN array(s) are of any size (10TB+) with lots of drives (8 or more) you should take another look at XFS. The version of RHEL you were using is eons old now, shipped with kernel 2.6.18 from 2003, and probably didn't have the 2007 fixes. Also, if you were running your / filesystem on XFS, that's not recommended, because most Linux distros don't have the right integration for doing so, as you learned. Run / on EXT2/3 and your data filesystems on XFS. / filesystems aren't write transaction heavy, and don't experience lots of random IO, merely log writes, so you get zero benefit from XFS, and you can experience integration headaches, depending on your distro.

XFS is currently the most heavily developed FS on the planet, more than BTRFS. It also has superior performance to all other filesystems with almost every workload. Even more so with the recent delay log option. When configured and managed properly, it is also one of the most robust filesystems available. It is very widely deployed, especially on massive DB servers, due to its superior performance with O_DIRECT.

If you had corruption issues, not merely trunc'd files, in the past two years, then you were running an OS without the 2007 patch.

Not necessarily. If this were a superscalar cock processing unit, with say, 2 cock processing pipelines, 2 cocks could be processed per cycle. In this case, a 5KHz superscalar cock processing unit could process 10,000 cocks in one second. Cock processing is all about efficiency.

You are just full of misinformation. Sudden downing of any _busy_ system due to power loss or panic is going to lead to data loss due to the buffer cache in Linux, and other factors. Full stop. Data loss != data corruption. XFS is absolutely not prone to "data corruption" under any circumstances. If you pull the plug on any system running any filesystem, you _will_ get truncated files and loss of data. This is not a function of the filesystem per se, but of the Linux buffer cache, and any write caches on the RAID card and/or disk drives themselves. Write all of your applications to fsync and the truncation after power loss problem will be less severe, but your performance will suck horribly. Again, this is equal for _all_ filesystems, not just XFS.

Maximum performance and maximum data integrity have always been mutually exclusive, and always will be. XFS kicks the crap out of all comers in write performance with delayed logging enabled. This means more data is held in RAM before flushing, more so than with the standard config, optimizing data transfer throughput and placement on disk decreasing fragmentation. Yes, this comes at a cost: If power drops, everything pending in the write buffer gets lost. This is why (real) datacenters are designed with many large and redundant UPSs and generators. However, again, data loss != data corruption. You claim above that XFS is prone to data corruption. That is horse shit.

Don't spread FUD. You don't use XFS, or at least any remotely recent (2007 on) version of it, so you don't really have a clue. And you obviously don't have an understanding of the function of the Linux VFS system and the buffer cache. As I stated, _any_ filesystem will suffer truncation and data loss if a busy system loses power when many writes are pending. If your apps make heavy use of fsync, you won't suffer as many truncations, but you will still have them occur, with ZFS, EXTx, BTRFS, etc. It _will_ happen with all of these. The only difference will be the severity, depending on how many write transactions were in flight at the time the plug was pulled.

I'm anxious to see you claim again, in your response to my points above, that ZFS is immune to truncation due to power loss. The only way this is possible is by disabling all caching, on the disk drives and controllers up through the entire software stack on the host, and using only applications that call fsync on every write operation. If this was actually done by anyone, disk write performance would drop to the point the system would be unusable. And this still wouldn't guarantee you wouldn't have a single file in the write buffer that gets truncated.

http://www.sgi.com/products/servers/altix/4000/

"SGI Altix 4700 incorporates the shared-memory NUMAflex® architecture, which simplifies software development, workload management and system administration. It supports up to 1024 cores under one instance of Linux and as much as 128TB of globally shared memory."

http://www.sgi.com/products/servers/altix/uv/

"Altix® UV scales to extraordinary levels-up to 2,048 cores (256 sockets) with architectural support to 262,144 cores (32,768 sockets). Support for up to 16TB of global shared memory in a single system image, enables Altix UV to remain highly efficient at scale for applications ranging from in-memory databases, to a diverse set of data and compute-intensive HPC applications."

"This infrastructure is supported by a complete HPC solution stack running on industry standard Linux® operating systems with the choice of Novell® SUSE Linux Enterprise Server or Red Hat® Enterprise Linux® Advanced Server operating systems."

Someone didn't do his research before spouting off. This is odd considering he simply regurgitated the information Red Hat fed him. I guess the press guy at Red Hat didn't know they already had their OS running on 16TB capable SGI monsters?

Apparently you're not from New Orleans, where you only go to jail if you don't/can't pay a bribe. Their drug unit took over the city's drug trade via forced bribes, if you will, forcing dealers to give up their business to cops to avoid arrest/jail time. At least, that's the way it used to be. The FBI crackdown some years ago may have cleaned the NOPD up just in time for Katrina.

I politely beg to differ haus. The .308 or .270 would both do a fine job here as well. Heck, we're talking about a puny pencil neck lawyer judge, right? One could easily and cost effectively purchase a .25 Jennings automatic in the North end of just about any U.S. city that would do the job, albeit at close range, and most of 'em already have the serial numbers filed off ta boot. Rich Texas judges usually live in two story white mansions with lots of shubbery, especially line'n the driveway--good hide'n spot for deploy'n that .25 Jennings auto and make'n a safe getaway. Silencers for 'em are pretty cheap, or you can easily make one yourself, being a country boy like me.

Say, uh, Slashdot don't log them there IP address thingies when people post, do they? If they do, umm, damn, uh.. I' never really done nut'n like 'at. I' just got a good imagination, folks tell me.

...and then there are the idiots all over the world who believe sticking "an" in front of words starting with hard consonants makes them...something. What? Smart? European? What? Who the fuck was taught in grade school or high school English class that you say "That is an chicken sitting on an egg."? It's not "an" chicken sitting on an egg it is "a" chicken sitting on an egg.

The English article "an" is to be used only in front of words beginning with vowel sounds when spoken verbally, such as "an honest politician" or "an English gentleman" or "an asshole!". The article "a" is to be used in front of words beginning with a hard consonant sound when spoken verbally, such as "she is a beautiful woman" or "is that a cocker spaniel pissing on your shoe?" or "you're a fucking jerk".

If one is using "an" in front of consonant words as some kind of post internet age form of "emphasis" I suggest it only makes one look goofy, uneducated, or striving too hard for some odd kind of attention. I still haven't figured out where this FUBAR'd execution of this English language article started. Was it Europe or here in the US? Or?

People I know personally, who know how to use this English article properly, have started using "an" in front of consonant words in print (emails) but never do it while speaking verbally through their lips. If I was sufficiently creatively inclined I'd put together a cartoon parody of this and stick it on YouTube to shame people into dropping this stupid plague that has sprung up all over the web.

No, the real problem is stupid Americans who allowed the fucking cell phone to become a status symbol. Recall the Motorola Razor? Same phenomenon as the iPhone but you didn't have Motorola Zealots in the purchasing crowd like with the Apple Zealots. I know 10 people who bought iPhones. One, only one, actually uses the features that would possibly justify the purchase. The other 9 bought it as a status symbol because they could afford it. Actually 3 really couldn't afford it but bought anyway. And these 3 are totally leveraged to hilt with big mortgage, car note, maxed credit cards, etc etc. They just had to have the iPhone to keep up with the Joneses.

Cell phones have ruined our social discourse in many ways. I got so fed up with the status symbol crap and having a cell glued to my ear that I canceled my contract one day 3 years ago and chucked the phone in the can. I regained aspects of my life that I'd forgotten I once had. I've since bought a prepaid phone that stays in the car for emergency use only, and I went back to carrying a pager, again, for emergencies only, whether family or business.

The freedom of no longer having to plug my head into and electronic gizmo 20 times a day can't be sufficiently described in words. I got life back, or at least most of it.

The announcement of this murderer's execution should be nothing more than a recording of the gun shots and the dirt hitting the box sent to the media.

What's the geolocation of 192.168.x.x?

It appears many of you have a horrible misunderstanding of how the VMware hypervisor works. One instance of VMware ESX can be installed on a single system image machine. SSI is defined by shared memory. There are two classes of systems with shared memory: SMP and NUMA. This 512 Atom machine is neither. It is a system of 512 individual machines cobbled together with very neat thermally and electrically efficient packaging. If one were to run VMware ESX on this machine, it would require 512 instances of the VMware ESX kernel. You could then assign guests to each single Atom machine, but you would not be able to schedule guests across/among the 512 machines. VMware, or any virtualization would be useless on this platform. That's not what it's designed for. To get efficient compute bound applications to run effectively on this machine would require something like MOSIX. For other applications such as web or webmail, you'd use a standard load balancer and a a cluster of 512 individual servers, just as is done now.

Again, this machine is not SMP and not NUMA. It is not a "server" in the traditional sense. In reality it is a _cluster_ in a box. The concept is nothing new, this company is rather later to the game, and is already beat in the efficiency department. One example beating it is the Sicortex SC5832 which houses 5832 64 bit MIPS cores running at 700MHz. The CPU die contains 6 MIPS64 cores, dual memory controllers, a PCIe interface, GigE interface, and a connection to the Kautz digraph network running at 2.5GB/s bidirectional to from each node. The system consumes less than 20KW for almost 6000 cores and over 8Tflops of floating point performance. This highly integrated cluster machine is designed for supercomputing, not web applications, though it could be used for such if desired.

http://www.sicortex.com/
http://www.sicortex.com/registration/download/265/1886/file/TechSummary_FINAL.pdf