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.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: Dan3 who wrote (91073)	2/2/2000 1:22:00 PM
From: Tenchusatsu	Respond to of 1575363

Dan, <simple file serving is moving to dedicated SAN components like NetApps, and that "servers" are more often now supporting applications with concurrent processes for each connection. Lots of task and context switches, so that even multi-megabyte caches can be overwhelmed.> Hence, the need for lots of memory bandwidth in servers. Actually, what I said before might be kind of misleading. Latency always matters, even for servers. But I'm saying that sustainable bandwidth affects latency much more in a server than anything else. Sustainable bandwidth affects latency much less on the desktop because desktop systems hardly ever saturate bandwidth. <Doesn't the use of one or more MRH units in series start taking a noticeable hit on performance? Aren't your total latencies approaching 300ns through the hub?> 300 nsec? You must be getting some false info regarding the latency of the MRH. The data I've seen suggests that the MRH-R component adds about 20 nsec to latency. That can make a real difference on the desktop, since average latencies of memory accesses are on the order of 80-90 nsec. But for servers, where average latencies are longer because of higher bandwidth utilization, the MRH-R doesn't make that much of a difference. Consider this: If you look at HotRail's 8-way Athlon chipset, you'd notice that the chipset will incur a LOT of latency because of the switched fabric topology. Even HotRail is estimating that the latency of their chipset will be 30 nsec longer than that of Intel's 450NX chipset (4-way Xeon). (I think it could be more, unless HotRail has a real high-performance solution for their central switch component.) But HotRail knows that sustainable bandwidth is much more important than latency. Tenchusatsu