zEnterprise and SAP: Perfect Together
I've seen some past criticism that IBM doesn't publish benchmarks for its mainframes. That's another "mainframe myth." IBM does publish benchmarks from time-to-time. Here's the latest example: a new world record for SAP core banking performance and scalability.
Benchmarks aren't useful at all unless they resemble some projected reality, in total. They're simply tools to try to understand how a system will behave when you put it to real, productive work. Unfortunately every business (and government agency) is unique, and when you're looking at a thoroughly mixed workload environment (IBM's zEnterprise), running a couple off-the-shelf tests isn't going to help much.
That said, this SAP benchmark is at least a better attempt, because it's trying to simulate what a real bank would do during the day, night, and across multiple channels and business functions. And the zEnterprise results are extremely impressive, even measuring SAP by itself. (IBM mainframes can and do of course run lots of applications on the same footprints.)
It's tough to benchmark "run most of my entire business" scenarios. But it's not necessarily impossible. I've worked with a lot of customers who pack up, go visit a benchmark center, and figure out how big their systems should be, how they should (or should not) architect their applications, etc. That's the best way to gain confidence about how well particular infrastructure(s) will support future business demands. (They can also test quality of service issues, such as how quickly they can fail over to a second data center, how much if any interruption there would be when upgrading software versions, and so on.)
Alternatively, if you've already got a mainframe, you're probably benchmarking all the time, perhaps without knowing it. You probably already know that your mainframe can handle the next merger or acquisition, and the next five applications, and then some. Maybe that'll require adding some capacity, maybe not, but the IBM mainframe inherently yields lots of performance-related data so you already know that you can deliver a quality outcome. After all, the easiest benchmark is the one you don't have to run. There are also affordable Capacity On Demand (COD) options for running special high-capacity tests on your own machine(s) — or to save your business if the programmers delivered some poor performing code that they can't fix right away.
Of course, if you have 150 million bank accounts, and you want to run SAP core banking, just go buy a couple zEnterprise machines from IBM (and a zBX or two). You've now got that benchmark, too.
|by Timothy Sipples||September 16, 2011 in DB2, Systems Technology |
TrackBack URL for this entry:
Listed below are links to weblogs that reference zEnterprise and SAP: Perfect Together:
At IBM, I used to do a lot of benchmarking in competitive situations - even once simulated a benchmark for the /360-85 before that hardware was fully operational. I found there were so many variables, so easy to tweak that results weren't an accurate representation of a real-life environment. There are too many factors affected performance - job mix, sequence it's presented, operator behavior, etc. Raw numbers in a controlled test don't mean much.
Example: Customer wanted total start/end time for running a batch of Fortran jobs. Treating them equally (as they had been run on a 7090) produced a time of about 110 minutes. Giving the CPU-bound jobs lower priority overlapped them with I/O-bound jobs and total time was 75 minutes. How the customer would run them in a real shop would depend on the SysProg's skills and attention.
Posted by: Ray Saunders | Sep 16, 2011 9:51:29 AM
Good information from Ray Saunders...
Liked it. :)
Posted by: Ganesh | Sep 16, 2011 10:32:05 AM
Thanks for that useful information..I am doing a course in mainframes and i had a greater information about the mainframes..
Posted by: Diamond Core Drill | Nov 24, 2011 1:25:30 AM
The comments to this entry are closed.