Re: It is almost certain now, INTEL will have 64bit x86 !!

From: Rick Jones (foo_at_bar.baz.invalid)
Date: 03/02/04 

Date: Tue, 02 Mar 2004 19:43:39 GMT

Andrew Harrison SUNUK Consultancy <Andrew_No.Harrison_No@nospamn.sun.com> wrote:
> Rick Jones wrote:
>> Andrew Harrison SUNUK Consultancy <Andrew_No.Harrison_No@nospamn.sun.com> wrote:

> What exactly don't you understand ? the MAXCPU configs on F15K's
> or the differences between this and the configuration of a 8400.

> Discussions about trading the 17 I/O slots (you need 1) for MAXCPU
> cards in F15K's are irrelevent in a discussion about the configuration
> restrictions of an AlphaServer 8400.

Tell you what - you can pick first - which would you like to be? The
pot? Or the Kettle? I'll take the other.

I didn't say the tradeoffs were identical, I said they sounded
similar. They are similar in that if you want the maximum CPU count
on the 15K, you have to have the minimum I/O count. If you want the
maximum I/O, you could not have the maximum CPUs on the 15K. I see
that as similar, perhaps you do not.

I saw what I guess is an opposite side of the coin similarity, or a
different form of tradoff if you like, in that on the FNNK's if you
want the maximum memory config, you _have_ to have CPUs.

>> You need to get in touch with the website maintainers and get them
>> to update the pages. Their cut and paste job from prevoius pages
>> has left MaxCPU references in place for the 25K.

> You could be right but since the maximum number of CPU's supported
> in the F25K is listed as 72 vs 106 for the F15K it should be obvious
> that the MAXCPU option has been dropped.

If it were all that obvious, I would have thought that the people
editing the web pages would have caught it :)

>> No, it was some marketroid taking 80% of that figure, and _still_
>> getting it wrong because they took the PCI slot count and not the
>> PCI bus count, and the sum of the PCI bus count is 13.5 GB/s.

> Unlike HP who just count the maximum bandwidth of the CELL board I/O
> bridge x 16 ! sounds just as scientific.

It is at least self-consistent as each of the up to 192 PCI-X I/O
Slots is its own PCI-X bus and so there is at least the theoretical
prospect of getting there.

"There are 192 I/O cards in the system, each with its own dedicated
I/O bus" from "Meet the HP Integrity Superdome servers A white paper
from Hewlett-Packard Company"

> Or claiming a maximum backplane bandwidth on the Integrity SuperDome
> and SuperDome of 64 GB/s despite only managing less than 1/2 of that
> on a non MPI Streams result.

>> Perhaps I make a math mistake somewhere, but can you tell me how it is
>> possible to have up to 21.5 GB/s of sustained I/O rate when the peak
>> marketing bandwidth of the I/O busses on the system aren't much more
>> than 1/2 that?
>>

> Because your maths isn't that good !

Very well, please correct it - how many distinct PCI busses, of what
frequency and width are there in the F25K?

>>>So what sort of I/O rate can you sustain through an Oracle DBMS on a
>>>SuperDome ?
>>
>>
>> No idea. I suppose that if someone were to poke around various
>> benchmark disclsoures some guesstimates of I/O rates for _those_
>> workloads could be made. To my knowledge, HP haven't done "just" an
>> "Oracle table scan" and published numbers. BTW, I would be curious to
>> know more of the details of that table scan, if you would please
>> provide a URL with the details that would be great.
>>

>> The "if you want the RAM you have to have the CPUs" tradeoff seems to
>> remain from the 15K.

> Sorry but again this is incorrect the tradeoff only existed in
> your imagination.

> A F4900-F25K system board supports 4 CPU's and 32 GB of RAM.
> A F4800-F15K system board supports 4 CPU's and 32 GB of RAM

I don't think the tradeoff was imaginary - what I claimed what that if
you wanted the RAM, you _have_ to have the CPUs.

So, specific question time again - Is it possible to have an F25K
system board with what Sun calls one CPU and access RAM in _all_ the
DIMM slots on that system board?

rick jones

BTW, I'm still waiting for you to give the URL with the details of
that Oracle table scan number you referenced. Stuff like how many
HBA's, number and type of storage, number and type of CPUs, quantity
of RAM, size of the table, size of the caches on the storage, that
sort of stuff. 

-- 
a wide gulf separates "what if" from "if only"
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com  but NOT BOTH...

Relevant Pages
Loading