Happy Holidays
Article 11/95
For many systems, the holiday season is a very busy time. Of course, your site is fully prepared for the expected peak in demand. Some of you don't really have enough of a variation of volume over Thanksgiving and Christmas for there to be any worry. Others, particularly retailers and POS transaction handlers, have to be prepared for a volume which may be 2 to 5 times their "normal" volume during the rest of the year.
This, of course, doesn't mean that the total transaction counts for the month or the week are 2-5x the usual. As successful OLTP managers, we know that we must design our systems for the peak half-hour. If a component of our systems hits saturation or starts up the "knee of the (response time) curve," the whole system may slow down and transactions may time out.
Let's review what we've done to be confident that we'll survive the holidays with very few calls to our pagers:
a) We know what we handled last year. We know the transaction volume that we handled each day of the season, and what our peak half-hour transactions were on those days.
b) We know what we'll be expected to handle this year. We've talked to our business partners and gotten their growth estimates, and since we've been tracking transaction growth during the year and comparing it with last year's, we have an idea of how accurate their and our estimates are.
c) We know what system resources were used to handle that peak half-hour traffic. We know how much CPU and memory were used/available during the peak half hour. We know the percent busy on each of the disks during the peak half-hour. We know the percent utilization of our communication lines during that peak half hour. We've upgraded any resource that may have been or might become a capacity bottleneck.
d) We know how hard the application software worked to peak last year. We not only looked at CPU figures on a per-process basis, but we used our performance monitoring software to help identify any servers or other processes that might be going over 50% active. ("Active" is a measure of the processing time, not just CPU time. It includes CPU time, time spent waiting on the disks, time spent waiting on comm, and time spent waiting on other processes.) We started additional copies of the more "active" processes so that they don't become a bottleneck.
e) We've taken a close look at the disk capacity, and we've done estimates based on the expected transaction volume and confirmed that we have enough space available for the expected transaction logs, work files, historical files, etc.
f) We've taken a look at out batch processing cycles, and done estimates to confirm that we can process the expected peak volume days within the window that we're given. We've also confirmed that we can ship this data out to the back-end or settlement systems in time to meet their deadlines.
g) We've told development and operations about the expected peak times, and asked them to schedule their non-essential activity accordingly.
Since we've done all that, we're going to enjoy our holiday.
| 
