Re: Informix ODS 7.24 on Sequent NUMA-Q

Laurent, Many thanks for your helpful comments. We are resuming our test tomorrow and I know we will benefit from your response. My faith in Sequent/Informix has been restored!! > Did this person told you why it wouldn't scale? Yes, because Informix was not NUMA aware, and this resulted in expensive accesses over the IQ link. >You do not mention scaling here. What is your reference point? Our only reference with regard to scaling was a decrease in performance with more users. 230 users ran at 45%, 80 users ran at 20%, combined (310) they ran at 75% (with reduced thruput) instead of 65%. This was blamed on Informix for not working well between the quads. We 'll probably shelf these observations and see what our changes do. >Did you actually run the very same application with the very same >user simulation on both Symmetry on NUMA-Q? No. My statement was based on my first hand knowledge of the applications and going through a stress test on both of them. >In this particular case, may I suggest that you increase the number >of Informix CPU VPs to 5 (4 on quad 1, 1 on quad 0). We never tried a fifth CPU VP. Rather we were told to contain the engine in quad 0, and set up a run queue to pull everything else to quad 1 (kind of like herding cats :-). We will change this tomorrow. It makes sense that quad 0 is left for the kernel. We'll follow your recommendations. Will it still be necessary to set up a run queue to pull processes away?? We currently have a cron job running rqadmin every minute. It somehow seems clumsy to do this. Also I read about NOAGE in the release notes. Is this of any concern?? >Do you have actual users typing or do you use a simulation software? We are using simulation software. Our thinktimes have been adjusted to enter an order every minute. (Although we want more!!) This is roughly the same in the other types of transactions. Each order entered performs about 25 validation queries, 10 inserts, and 1 update. 120 of 310 simulated users are entering orders. It may be hard to tell now, but we may need more than 2 quads. We'll know more tomorrow. >You claim here that you max out 2 quads with 310 users, they must be pretty busy :) Is 75% considered maxing out?? I thought that 310@75% meant that we could do 400 before we got into trouble. 450@40% is our goal. How many users will run on a machine thats meets this goal. We thought 1000. (or 500 a quad). We have 550 users now IRL and are growing at 23% a year!! I'll will follow up with our results tomorrow. Thanks again, Dave