Rob van der Heij commented: > On 8/30/06, John Campbell <[EMAIL PROTECTED]> wrote: > > > The hell of it is that I *don't* have a zSeries-- or even an > > instance within z/VM-- where I can run Linux, so I can only do > > some speculation. > > Yes, and that is where it makes a difference. On discrete servers it > may make sense to tune your configuration for maximum single server > performance. Any resources that you don't use are wasted anyway, and > there's nothing wrong to increase CPU usage from 10% to 20% if that > buys you 10% more throughput. Your tuning is not complicated by other > things happening on the same machine (although a shared disk system > like a SAN is going to impact your tuning if the bottleneck is other > than on your own doorstep). > > With Linux on z/VM the tuning objective often is to achieve lowest > cost per transaction (as long as you meet the required response time). > It is very rare that you need to size your virtual machine to consume > all available resources of the entire machine (like CPU or I/O). In > most cases the z/VM system will also run other virtual machines at the > same time and normally your objective is to have all of them make some > progress rather than let one virtual machine run away with the system. > Oversized systems then require extra monitoring to prevent them from > taking all that you gave them.
Well, I commented to Ken that this is because you're not merely tuning for "best throughput" but you have to balance it with "being a good neighbor", so there are two targets you have to consider. Now if only I had an opportunity to amass *real* experience. I was once pretty good at working w/ standalone Unix systems and even did some performance analysis on mainframes... but not of zSeries, but on Sperry 1100 systems. It's been a long time since I wore _that_ hat. (laughs) I have *way* too much "useless knowledge" stowed away in my head. Well, at least I'm pretty good w/ pSeries boxes... -------------------- John R. Campbell, Speaker to Machines (GNUrd) (813) 356-5322 (t/l 697) Adsumo ergo raptus sum MacOS X: Because making Unix user-friendly was easier than debugging Windows. Red Hat Certified Engineer (#803004680310286) ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
