Mainframe specialty processors: Do they really save money?
IBM has been pushing mainframe specialty processors — the IFL, zAAP, and zIIP — as a way for mainframers to save money. But do they really?
Trevor Eddolls at the Mainframe Update blog writes about this very topic, calling it a “bit of a swings and roundabouts situation:”
An organization needs to pay for a specialty processor, and it needs to buy software that can make use of the specialty processor, and then it can start saving money – I hear the sound of Excel Pivot tables being used to present the result of what can be quite a complicated calculation. But is there any software that can make use of this hardware?
Eddolls goes on to mention DB2 as one prime workload able to be moved to the zIIP, and there are some other major software vendors out there — CA, BMC, DataDirect and Neon Enterprise Software — that are looking to offer customers the opportunity of running their software on zIIP or zAAP.
IFL, the Linux processor, has been a success for the most part, mainly because IBM has sold it — and customers have bought it — as a way to consolidate hundreds, if not thousands, of Linux instances onto a single box. That has potential savings implications in floor space, power, systems management, and staff.
Selling the zAAP for Java and zIIP for data applications hasn’t been as easy. Why? The consolidation attraction isn’t there for one, and the benefits are more subtle and indirect. The case for the zAAP and zIIP is that you can save software licensing costs by getting workloads off the central processors to the zIIP and zAAP, where workloads don’t count against the million service units (MSU) rating of the box that is used to calculate software licensing costs.
So you might be able to save in software licensing costs, but the processors themselves cost about $100,000. In talking to users, it seemed to me that the benefit of the zIIP and zAAP was more indirect. By taking workloads to those processors, you can free up room on the central processors. That’s what might matter the most.
So instead of having to buy a new mainframe, you can just buy one or two of these zAAPs and/or zIIPs. The real savings comes not from the reduced software licensing costs as much as it does from being able to buy a six-figure specialty engine instead of a new seven- or eight-figure mainframe.
Posted: May 20th, 2008 under Mainframe servers, Modern mainframes.
I’ll agree that the initial $100k might seem a bit steep but consider that the true cost of just one Linux server is around $10k. You only need to virtualize 10 servers to break even. If you have a few hundred lying around, that could result in some pretty hefty savings.
Don’t forget that IBM has significantly dropped the price for z/VM to be about the cost of a single Linux server… again about $10k for a single license that can be used to virtualize literally thousands of Linux servers on a single z box.
I’m not saying to abandon servers you’ve already paid for or that are in the middle of a 3 to 5 year lease. Put new workloads on the mainframe specialty servers instead of buying or renting new servers and let the old technology roll off on schedule. You’ll be amazed at how quickly those expensive specialty processors can pay for themselves.
Not only that, but you can still run your vintage 1968 COBOL apps on the same box. And you get all of this without having to increase your datacenter rack space, power and cooling utilization, property taxes, etc.
Sounds like a winner to me!
Comment by An old mainframer — May 21, 2008 @ 11:39 am
Yes, I said property taxes in my previous post. Most folks don’t realize that, depending on your datacenter’s location, your company may be paying property taxes on all that equipment that’s sprawling across your raised floor.
If you don’t have to add equipment to virtualize more Linux servers, you don’t have to pay extra taxes.
Hmmm…
Comment by An old mainframer — May 21, 2008 @ 11:42 am
With the upgrade we are inthe middle of, we anticipate the reduced software costs from ISV’s as a result of buying a smaller than predicted machine - the difference going to the ZIIP-
of roughly 2.5 Million dollars. The new specialty processors are really quite impressive from a cost/benefit perspective. The additional benefits once we get more experience will just sweeten the pie even more.
Comment by Roger Webster — May 27, 2008 @ 2:22 pm
Can anybody explain to me the performance-related benefits of zAAP?…All I read in the definitions is “save software costs”…I’ve monitored our PROD Websphere JVM using Wily (AvgServletRespTime) and I don’t observe, upon a gross examination, much of “user experience” boost here? The SYSPROG manual for WLM says the following which I don’t get (focus on IP stack and connectivity):
“…The benefits of using a zAAP processor to execute Java code are saving CPU cycles in IP
stacks and in the firewalls due to the connectivity simplification…”
Can anybody comment on the expectected (or not ) performance boost?
Comment by Michael Dean — June 26, 2008 @ 3:40 pm