XGrid Info?

[illumin8]

Re: Re: Sun GridEngine

Originally posted by marshy
Hey, if you're willing to sell me dual processor V210's with 4GB RAM for less than $3k I'll start up a reseller business tomorrow... ;-)

In fact, if you look at your employer's web-site (here -> http://store.sun.com/catalog/doc/BrowsePage.jhtml?catid=100054) you'll find that it costs $5795.00!!

Still not expensive (for what it is), but definitely not the bargain you make it out to be....

You are correct. It's actually the 1 proc @ 1ghz. with 512MB of RAM version that is under $3000. In the marketing presentation I watched they compared it side by side with a Dell PowerEdge server and the PowerEdge came in at something like $3500 for a similar configuration, before even adding the Windows tax.

Either way it's a good deal for a real Unix system. We have had to drop our prices tremendously to compete with the Wintel/Lintel boxes that are out there.

 

[jamilecrire]

Originally posted by MrMacman
Hm... may be good for buisness, but you never know do you?

As a owner of my own business the only other platform I'm considering is OS X Server (we currently run Linux and unfortunately have 1 windows server). My company will be adding more servers in the next 6-12 months but I'm willing to wait a little longer for apple to release 970 based XServes. I have no need for XRaid or XGrid.

The ONLY two things holding me back on pulling the trigger are Oracle is in Release Candidate Stage (2 which it has been for a long time) and I need a 64bit platform for database tasks. If Apple doesn't show me something in the next 12-18 months I'll be stuck with running Oracle on either Xeon based (yes I know 32bit) Linux systems or plunk down insane amounts for an UltraSPARC based system (I currenly have one). I guess the new AMD Opteron (spelling?) is an option but it's the same as OS X Server. No Oracle, although IBM said DB2 is currently being ported to a native 64bit Opteron version.

Oh well, I guess the focus of this novel is 970 based XServes, and the associated hardware (XGrid/XRaid) are excellent for business. Look at how much a dual 1.05GHz UltraSPARC system is...

Edit: I forgot Oracle is making an Opteron version of 9i, it's in Developer Release Candidate Stage 2 as well. Guess Linux/Opteron may be my choice. Come on Apple, give me a reason to question my decision.

 

[phrantic]

Originally posted by ennerseed
doh- forgot to actually respond.

ppc chip's, unlike other's, processing power doesn't start a decling the more processors you add. in fact tests show there maybe no end to the amount of ppc processors that can be added.
where on other processors the power from adding starts to slow down, until some point it really won't make it any faster to add more processors.

Unfortunately, without you linking to some supporting documents, I'm calling this post BS. Parallel processing problems such as Cache Coherency, the bandwidth limitations between processors, and latency are three BIG drawbacks that will forever keep processors (no matter who makes them or what the processor architecture) from being added in large numbers. 4 processors on one card (in an SMP machine like the Sun 15K http://www.sun.com/servers/highend/sunfire15k/specs.xml ) is about all you can manage before you have to start interconnecting cards via huge crossbar switches and whatnot. But latency soon becomes a big factor (as well as heat and power consumption) in how large you scale machines with cards of 4 cpu's. That's why the Sun 15K tops out at 106 processors.

Please defend yourself or quit spreading FUD. PPC is not the end all be all of processors. It does not defy the limitations of physics and the speed of light.

 

[phrantic]

But I guess you were more or less quoting this

After running a series of numerically-intensive trials on a 33-node XServe cluster, we were able to achieve over 1/5 TeraFlop on certain problems. These results were very repeatable. No evidence of an intrinsic limit to the size of a Macintosh-based cluster could be found. Building on a previous result using 76 Power Macs at USC, this finding is further evidence that Macintosh-based clusters are capable of excellent scalability in performance.

Fine, they found "no evidence of an intrinsic limit to the size of a Mac-based cluster," but being mac-based or not, given their tests and results, they wouldn't find scaling problems among other platforms. They failed to acknowledge the effects of the interconnect (just 100BaseT?) and the fact that fractal generation is "Embarrassingly Parallel". If clusters can have "no intrinsic limit" and do all their message passing over 100BaseT, they why do people like myrinet http://www.myri.com/ and Quadrics http://www.quadrics.com/ exist and provide incredibly high speed interconnects? Or check out http://www.top500.org/lists/2002/11/ to see the lists of the top 500 cluster (usually) computers in existence. Take note of the number of nodes and processors, the interconnects, and then think if their tests on a 33 node Xserve cluster truly shows a _lack_ of an intrinsic limit to PPC clustering. That's a pretty bold claim based on a tiny cluster (in comparison) running an embarressingly parallel problem.

 

[Snowy_River]

Originally posted by phrantic
But I guess you were more or less quoting this


Fine, they found "no evidence of an intrinsic limit to the size of a Mac-based cluster," but being mac-based or not, given their tests and results, they wouldn't find scaling problems among other platforms. They failed to acknowledge the effects of the interconnect (just 100BaseT?) and the fact that fractal generation is "Embarrassingly Parallel". If clusters can have "no intrinsic limit" and do all their message passing over 100BaseT, they why do people like myrinet http://www.myri.com/ and Quadrics http://www.quadrics.com/ exist and provide incredibly high speed interconnects? Or check out http://www.top500.org/lists/2002/11/ to see the lists of the top 500 cluster (usually) computers in existence. Take note of the number of nodes and processors, the interconnects, and then think if their tests on a 33 node Xserve cluster truly shows a _lack_ of an intrinsic limit to PPC clustering. That's a pretty bold claim based on a tiny cluster (in comparison) running an embarressingly parallel problem.

Just thought I'd clear up a point here, not that I'm in any way intending to defend the previous poster's position, as I don't know enough about the XServe to make any kind of judgement...

But, the XServe does not have 100Base-T ethernet connections. It, in fact, has Gigabit (1000Base-T) ethernet connections.

Just thought I'd clear that up.

Uh... one other thing, I think...

Isn't the point of a cluster to run very, very parallel operations (whether 'embarassingly' so or not)? The instances where I've most often seen clusters used are areas where the exact same set of calculations needs to be run over a wide variety of data, or something similar (as in high end molecular modeling, or super-collider data analysis). If that's not as parallel as fractal calculations, then I'm not sure what is. This is the kind of application that I'm excited about seeing Apple move into the server industry for.