We are in the process of upgrading hardware on our two mail servers (both support about 800 users each).
Old boxes (IBM xSeries360 Quad) with 4 1.9GHz/400MHz processors - Xeon
New boxes (IBM xSeries346 Dual) with 2 3.8Ghz/800Mhz processors - Xeon
My network guys tell me that the new server will be much faster, and I understand the basics of why - though I have not found any way to validate that claim. Anyone have any details or comments to help me finalize this decision with more then a good guess?
Your licensing costs for the new one will be cut in half, and the faster bus and (presumably) better i/o subsystem in the new boxes should give you better performance. The place to go to validate this is www.notesbench.org.
A couple years ago chip makers were extolling the virtues of “hyper-threading”, a high-tech term for simultaneous multi-threading (SMT). An important item they didn’t always mention was that in order to take advantage of this, applications have to be multithreaded as well. So the higher multithreading efficiency of an application, the faster it goes.
The actual history goes something like: simultaneous multiprocessing (SMP) or dual processors; multithreading, super-threading -or time-slice multithreading, and then hyperthreading (SMT).
Without a huge technical dump here, any advantages that multiprocessing (quad) brings, SMT brings way more in efficiency, particularly in the way of context-changes (the actual way a CPU handles more than one thread). There is no replacement for more CPU’s (deeper blue), but an SMT configured system with duals that are at twice the clock speed of your quads will (should) provide huge increases in raw processing - both in speed and multiple processes.
Further, IBM has indicated in some of its R7 previews that it takes advantage multithreading capabilites of the new CPU’s (they have some actual test data posted somewhere to prove it). But even running the same versions of Notes that you have now, I suspect that server software like Domino will see some impressive increases in performance.
I know this is vague and not detail-heavy for your purposes. It’s mainly just my ramblings while I pretend to be busy doing processing-thirsty development work. (and I wa trying to keep this brief)
Thanks. I am now more confident that the new Dual boxes will be better.
To the question of re-configuring tasks… you say if I had 3 update tasks, to throttle that back? why is that? Should I be matching tasks up to the # of physical processors?
Update tasks are usually bottlenecked at the CPU (though this varies case-by-case), so having multiple Update tasks running per-CPU causes that CPU to be context switching the Update task off-and-on, wasting clock cycles that could be spent running Update in the first place. So a 2-cpu box running 8 Update tasks will spend tons of time just moving the Update tasks on/off CPU as opposed to simply running them and getting real work done.
If you had 4 Update tasks before, I’d run 2. If you had 3, I’d seriously look at 1.
I’d test it with 1 in any case, as it might be plenty fast for your needs, and if you’re running multiple Update tasks you run the risk of 2 update tasks updating the same view at the same time, which reduces efficiency (both in CPU and precious drive I/O).
I agree with all of the above, which jibes with my experience. The hyperthreading combined with the increased bus will blow the old boxes away. Also, make sure to get as much memory as you can afford, it’s cheap now.