In article <8d6f4402.0307160034.644382c8@posting.google.com>, Johan
Manning <johan.manning@network.se> writes

John, can you specify your setup a little bit?
cuz im running two servers with dual 2.8 xeons and im not getting
anywhere near you results. Im running windows server 2003, cli 3.03
with standard settings for affinity and priority and im doing about
5h01m average based on 771 wu's. Number are from running 4 versions at
the same time ofcourse.

Sorry for the long answer to your simple request, but ......

Let me back track slightly before outlining the spec ...

The figures I produced were under 2K server during the burn in (and more
play after), and over a few WUs (75) as we experiment. The dual proc is
still largely with my Box builder, and the system is very new to both of
us. The WU figures I quoted were the ones he let me have.

Before I can fully lay my hands on the system, within my small SETI
farm, I agreed he could explore the system under different Bill G OSs. I
assume I will get the full system mid next week. I can then give it
welly over a month, and plan the following schedule -

        1 instance of the CLI, plus Driver & Spy. No processor affinity
        2 instances of the CLI, plus Driver/Spy and proc affinity set
        3 instances of the CLI, etc
        then all 4 running.

        NOTE: These trials will be under Win2K Pro, which I know do not 
        recognise HT. I want to see the impact of 2 things -

        1.      Slow down due to 2 instances of the CLI from shared
                RAM and Bus
        2.      Further slowing on a 512K L2 processor set.

        NOTE: I have run a dual P3 933 MHz server for a couple of years.
        My single proc P3 peer-to-peer currently runs the CLI 3.03 at 
        7.1-7.15 hours per WU (averaged over 2000+ WUs). The dual proc
        runs the same WUs in 7.5-7.55 hours (on well over 6K WUs). I am 
        assuming the same effect will be present in the Xeon dual proc, 
        despite the dual channel RAM. What I am hoping is this will help 
        the 2 instances of the CLI, and be hindered in running 4.

Interestingly, we found that the system crashed when HT was switched off
in the BIOS under Win2K Pro. This is the OS I want installed (not XP
Pro), as the other peer-to-peer Boxes on my network run it.

Why my box builder is experimenting is he is now supplying business Xeon
systems, and wants to push some of the parameters. I am relaxed about
when I finally get my box, where as businesses insist the contract time
is important.

Now the dual Xeon system spec ...

        Intel SE7505 UB2 server/Workstation MoBo
        2 Xeon 2.8 Ghz procs (512K L2 & 533 MHz quad pumped)
        1 GB DDR PC2700 RAM
        Serial ATA 80 GB HDD
        Intel SATA Interface
        ATI Rage XL Graphics integrated (16 Mb)
        Standard 2 channel sound card, for Voice Recognition
        2x Intel Pro/100+ server network connectors
        V92 internal modem
        52x CD Rom Drive
        Sony DVD-RW all standards
        Floppy
        A couple of EP-80 standard quality speakers
        Intel SC5250-E Server Case
        Win 2K Pro as the OS

        NOTE: I am in a broadband deficient part of the UK, and an
        area, without taking up Satellite Broadband, that is highly
        unlikely to provide it. I have the further problem I am 8 km
        (5 miles) from the exchange, and UK ADSL only reaches 5.5 km 
        (3.4 miles). However, within the next 3 months a local company
        is specifically supporting an ISP service based o 8.2.11g WiFi
        - 54 Mbs raw 2 way data exchange.

I will post the accurate figures I get for these trials for interest.
So, if your experience is applicable (most likely) then I will report WU
crunching times almost double those I originally managed.

The reason I am hopeful of, at least, achieving the better end of CLI
crunch times (pleased if below 4 hours) is my experience with my dual
and single proc P3s. These seem to be at least 1.0-1.5 hours faster than
the averages quoted by others posting to this forum. Also when looking
at the figures quoted at Team Ars Technica, etc.

I hope this was not too long ; )

John
-- 
John Clark