Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?

To: Tenchusatsu who wrote (133735)	4/30/2001 5:19:56 PM
From: pgerassi	Read Replies (1) \| Respond to of 186894

Dear Tench: Doing an 8 way UMA SMP is much harder than a 8 way NUMA SMP. I agree with that. But you are assuming they will attempt a UMA when their diagram of last year was a NUMA where cache coherency only applies between the two CPUs connected to a NB. Beowulf systems show that connection via Non Uniform Memory Access (differentiating between local Uniform Memory Access memory (the memory bus attached to the particular NB) and the remote memory attached to another NB's memory bus make cache coherency unnecessary between local CPUs and remote CPUs) is relatively easy, well understood, and can be done without huge effort. NUMA MMP systems appear far more quickly than UMA MMP systems. They are also much larger, scale better, and have higher price/performance. Just look at computers like HyperCube. I do have experience with SMP in all three types, UMA, NUMA, and loosely coupled (multiple systems coupled with a network of P2P links). The problems with a NUMA style are much overshadowed by compatability testing. In servers, the testing is much more involved and bugs have more significance in 24/7 lights out operations. Princeton even lost a router bridge for years. It worked but they couldn't find it (it was discovered behind drywall after painstaking cable following). Pete