Free Republic
Browse · Search
General/Chat
Topics · Post Article

To: ShadowAce

The bottle neck in computer data management has never really been about processor speed. But about ‘retrieval and storage’ speed.

You can have the fastest processor in the world, but if your data stream runs at the speed of a 386, then the processor is going to be waiting a long time before it can even begin any calculations or data handling.

Even ‘solid state’ drives aren’t adequate enough,in terms of speed, and there useful life span, and replacement cost, isn’t much of a help either.

They need to develop a better way to access the data.


3 posted on 11/30/2011 6:00:50 AM PST by Bigh4u2 (Denial is the first requirement to be a liberal)
[ Post Reply | Private Reply | To 1 | View Replies ]


To: Bigh4u2

there = their.

:(


4 posted on 11/30/2011 6:01:37 AM PST by Bigh4u2 (Denial is the first requirement to be a liberal)
[ Post Reply | Private Reply | To 3 | View Replies ]

To: Bigh4u2

When you move into the HPC arena storage speed can be less important than interconnect speed. Typcial algorithms are distributed across 10’s of thousands of compute nodes and each subdivided part of the algorithm has to share intermediate results with other parts running on many other machines. The message rates can be horrendous and the network latencies can become the bottlenecks.

And actually, processor speed was the bottleneck in many applications. During the 70’s and 80’s, as a mainframe vendor (we were one of the 7 dwarfs) our performance department always couched their reports in terms of processor cycles. I constantly tried to get them to pay more attention to I/O (input/output) to no avail. They once produced an “I/O performance” report, but when I read the details the figure of merit was the number of machine instructions per I/O!


9 posted on 11/30/2011 6:30:33 AM PST by the_Watchman
[ Post Reply | Private Reply | To 3 | View Replies ]

To: Bigh4u2

What you’re really describing is “message rate”. The fastest message rates are obtained either in highly-proprietary interconnects from the likes of Cray or IBM (with their Blue Gene systems)in “capability”-level systems; commercial OTS solutions would clearly be Infiniband (QDR, or quad data rate, =40Gb/second). That’s pretty smokin’.

Disclaimer: This is my business; my field....and has been for years. I currently work with one of only two remaining Infiniband suppliers in the world.

The author of this piece clearly isn’t an HPC type, but that’s ok. I was at SC11 in Seattle (I go to SC every year). Great show as usual, sucky weather. Frankly, there really were no show-stopping announcements at the show this year. Still, it’s always a lot of fun.


15 posted on 11/30/2011 11:28:15 AM PST by RightOnline
[ Post Reply | Private Reply | To 3 | View Replies ]

Free Republic
Browse · Search
General/Chat
Topics · Post Article


FreeRepublic, LLC, PO BOX 9771, FRESNO, CA 93794
FreeRepublic.com is powered by software copyright 2000-2008 John Robinson