Mike, Re: "Are you saying :
a. Merced chip performance is as fast as
mainframes,
OR ...
b. Merced is targeted at mainframe market"
re a., Merced chip performance, on a single chip basis is probably going to rival that of a single mainframe CPU chip. However, mainframes scale excellently up to 12 ways in tightly coupled SMP configurations, so to get the real big MIPS, the Fortune 100 companies, etc., buy the highest configurations (most CPU's, memory, channels, etc.). I think Paul said Merced will scale up to 256 processors. However, the scalability of NT is well known to need a lot of improvement, so it will be one of the big bottlenecks in Merced based systems, in terms of performance, to get to rivaling mainframes. It doesn't matter how many chips you put together, if the operating system doesn't get the most out them, performance won't come up. It's sort of like a saturation phenomenon, or an asymptotic thing.
re b., you are absolutely right. High (extremely high) I/O bandwidth that is managed by IOP's that just get kicked off by the CPU, and then take over independent of the CPU is key (you mentioned this, in part). And, yes, all the other things you say in "b" are important to be able to do if you really want to call yourself a mainframe." And, there are others I've put into previous posts.
Re: "Are you aware of Intel working on these
issues?"
Intel is slowly building up their redundancy and other RAS capabilities with offerings like Failover and Landesk, and features like ECC on caches. In the I/O area, they do have independent IOP's available. However, I don't know if they are working on things like data integrity testing inside the CPU chip, super high I/O bandwidth (that's more in the SCSI, Fibre Channel, fast ethernet, OSA arenas, anyway and I don't know if Intel has sights on those). Security, ala hardware encryption with tamper proof hardware? Don't know. Only time will tell on all of this.
Re: "Or is Intel targeting the
SUN/SiliconGraphics/UNIX workstation market
with a super chip? Thanks for the info."
I don't work for Intel, so I am not privy to their targets with Merced. But, yes, I think it is obvious that they will be going after Sun, SGI, NCR, Bull, Unisys, and anyone else making mid to high end servers. The Sun Starfire series, that scales up to 64 way Sparcs, would be an obvious target, to me. BTW, the biggest Starfire config. goes for about $3 Million. And, they are selling very well for Sun (not just the big Mama, but, probably mostly medium to upper medium configs). Lucrative business there. SGI is hurting big time, and they probably will get hurt more by Merced. Actually, they have announced that they will be building systems around Merced. Trouble is, most customers will not think of SGI as quickly as they will think of HP, Dell, Compaq, IBM, DEC...when they want to buy NT servers. So, I don't know about the future of SGI, or Sun, for that matter. Sun is the last holdout re building with Merced, although they said they'd support getting Solaris up on Merced. There is no love lost, of course, between McNealy and "the Evil Empire", but he might just go down in flames with his ego/refusal to recognize what everyone else has.
So, yes, to people really in the know, it takes a lot more than just MIPS, or GIPS, in a machine for them to consider it a mainframe. With Merced, Intel is working up the foodchain. Historically, they have gone from swallowing minnows, to guppies, to trout, to barracuda, to Tuna, to small sharks, to medium sharks (the latter four with Merced). But, they have a ways to go before they take on the big Sharks.
Tony
p.s. Poor AMD and Cyrix have zero chance at any of the above. Boo hoo. |
