zIIPs and zAAPs

(Originally posted 2006-01-24.)

Today we announced the zIIP specialty engine for System z9 109. When I say we I had nothing to do with it, of course. 🙂 I’m in Global Services and we don’t make processors – we only tune them. 🙂

It seems to me zIIPs are going to be more significant than zAAPs. zAAPs (sometimes known as IFAs) are offload engines for java – and so are relevant to new-fangled workloads. I do know customers who have installed them, and others who have trialed them. Which reminds me I need to catch up with the two big trials I’m aware of. Here’s a link to one of them – with 13 zAAPs. But most customers don’t run tons of java on z/OS yet.

zIIPs, on the other hand, (sometimes known as SUPs) are going to be of interest to installations who:

1. Run DDF work against DB2. These include Peoplesoft, Siebel and SAP R/3. (The latter was rearchitected to use DDF for DB2 Version 8.) Also quite a few home-grown applications, including the dreaded Excel spreadsheet. 🙂
2. Do lots of IBM DB2 Utilities work.
3. Use DB2 Star Schema processing. These are most common in Data Warehouses, such as SAP BW.

Now, in my social circle are lots of customers doing 1. , quite a few doing 2. (but perhaps not burning much CPU doing it), and a few doing 3. (SAP BW users, mainly). So I conclude this is going to be really quite popular.

I have been in conversation with a number of developers and under the covers this looks a lot like zAAPs. Certainly in RMF you should expect to see the SMF 70 record (and reporting) treating them like zAAPs. But they do have their own processor pool (Pool 6) and their own sample counts, etc. So everything we learned about zAAPs appears to hold true for zIIPs.

And our code in PMMVS for processor analysis is going to have to evolve. Luckily we did a lot of work on this over the past two years. So this should be a natural extension of what we’ve already done.

And, naturally, I feel pretty good about assessing the scope for zIIPs – given the right data.

Finally, during the conversations on zIIPs I got to have some pretty nice discussions on things we could add to our instrumentation for processors in general. 🙂 That’s my bit of influencing Development done for the year. 🙂 But I can’t tell you about these neat enhancements right now. 😦 And anyway they will look better with some sample data than they would if I tried to describe them to you.