POWER6, Workload Partitions and Mobility

In the last month, I should have written a number of blog posts on our latest product announcements, instead I’ve been really busy. I have been spending most of my time on a root and branch review of what I’m calling platform management. I was also in Orlando for the IBM IMPACT 2007 conference briefing press and analysts on the announcements, but spent much of my time on a couple of key POWER7 topics.

More on that platform management later, but for the record here is a quick summary of some key part od the announcement, and it’s a doozy(1).

POWER6

This week saw the much speculated ECLIPZ, or POWER6™, more correctly the IBM System p™ 570 server, based on innovative IBM POWER6™ processor technology. The “server” can go from a 2- to 16-core configuration, each core supporting 2-threads.

Interestingly one of the neat features is the ability to switch the processor into low power mode while still running. In most cases this will save up to 50% of the electricity it consumes with little effect or degradation seen by individual applications.

Decimal Floating Point

Much misunderstood, while the focus has been on the power, energy consumption and throughput and stunning benchmark results, one of the more interesting and valuable features is the addition of Decimal Floating Point instructions in the Power Architecure.

This is the first time IBM has put base-10 decimal instructions, and it is the result of probably four or five years concerted vision and effort of IBM Fellow, and fellow Brit’. Mike Cowlishaw. Mike realised that as computer architectures moved to 64-bit and possibly larger, commercial business programs would potentially suffer inconsistent and a best potential rounding errors in decimal arithmetic during the conversion from decimal to binary and back again.

This was especially true in business languages such as COBOL and his own procedures language REXX, and others. Mike was able to marshal through various standards organisations a platform neutral standard implementation and with the launch of Power6 it is now in hardware, and a number of compilers, middleware packages and databases have been updated to use these new instructions. Oh yes, becuase the new instructions are in pure hardware they also run faster and improve performance of applications that use them, but thats a secondary benefit!

Mobility

Live Partition Mobility will migrate a running partition and its’ workload to another partition on another IBM System p server. This suppoirt will include AIX and Linux, and be extended to i5/OS in time. It has some great applications, not just an availability feature for when you need to to do hardware maintenance, or restarts etc. but also potentially a key feature in organisations that are heavily focussed on energy management.

Say you’ve got three Power servers, over the weekend and during holiday periods the workloads on two of them is very lightly used. Instead of shutting the systems down, or building a complex cluster based application where you can shutdown nodes within the system, you simply use Live Partition Migration to migrate all the workloads running on the two lightly used servers onto the 3rd. Once complete, usually within 15-minutes, you can shutdown and power off the two servers you are no longer using.

Prior to the workload useage returning to normal, you just power on both servers and use Live Partition Migration to move the workloads back. No system outages, no file system recoverys.

Workload Partitions

Part of an upcoming formal announcement of AIX 6 will be a feature called Workload Partitions. Yes, you might say this is similar to Sun Solaris Container function, but I wouldn’t!

Workload Partitions are a key part of a development effort to overhall AIX over the next few years and move it much more into a federated, services based model. Workload partitions allow you to quikly and easily start multiple instances of workloads within a single copy of the AIX operating system. Each Workload Partition has memroy and file space seperation from each other and share commonAIX libraries.

One reason why AIX Workload Partitions will become increasingly important is to take up the complexity of running applications in a multi-threaded, multi-core processor. Along with Logical Partitioning and Micro-Partitioning, it will allow users and programmers to relatively seemlessly use the ever increasing power and throughput over these systems without resorting to complex multi-threaded programming, queue and lock management etc.

Application Mobility

Also part of the AIX 6 announcement will be Application Mobility. This provides the mobility between AIX instances on the same and different physical Power based servers of work running in Workload Partitions. In the short-term this will be useful in operational availability and energy management. Longer term it will become a key feature of an automated, fedrated, workload managed server infrastructure for AIX.

So, thats a wrap for now. I had hoped to write more, and more timeloy as well as more detail, but work and immenent vacation meant I was left frantically typing this a couple of hours before leaving for the airport.

(1) A unique or strikingly different one of its kind