In search of partitioning

In his Enterprise Architecture: Virtualization and Management by Magazine blog post, James McGovern muses on mainframe virtualization leadership and if the likes of James Governor and the 451 Group will start blogging about it. He also wonders if “IBM mainframes would make a better participant in a grid architecture than Sun, Dell or HP?”

It’s not clear where the link is to management by magazine, but the blogsphere is certainly a funny old world. Partitioning and virtualization is taking off in a big way and a few short clicks this is all back in focus.

Back in late 2001 and early 2002 I was involved in a public spat with then SUN Microsystems Chief Competitive Officer Shahin Khan over a piece he’d authored on running Linux on mainframes, and IBM’s commitment to both mainframe virtualization and Linux.

Given I’d contributed the strategy and much of the impetus on IBM’s mainframe Linux initiative back in 1999 and in a 2000 Redbook, it was certainly a little galling to see Sun trying to dis’ both technologies. Most of the written submissions were published on the LinuxGram, an online subscription news service, sadly I’m no longer a subscriber, and the content is not generally available so linking is not possible. Sun have removed the original paper from their website as well.

So why bring this up? It seems James McGoverns’ post was triggered by the reports on InformationWeek about Nationwides’ virtualization of some 700-servers. The article doesn’t contain much detail but gives some useful numbers. I’m delighted to see that organisations are getting some real business value from this technology now, and this is by no means the only one I’ve heard of. But the game has moved on, I’m certainly more interested and excited about the forthcoming IBM System p servers and their function in the partitioning arena.

I’ve enjoyed catching up with, and finding out about some of or customers and their current exploitation of System p logical partitioning. We’ve got a number that are driving the partitioning to its current limit, running 70-80 real production partitions, a mix of both AIX and Linux on a single physical machine/server. This year we will be lifting of those limits and shipping some entirely new functions which should allow these and other organisations to save even more money and deliver entirely new business value.

Oh yes, and what about the blogsphere, Governor, 451 Group and Grid ?
By a strange coincidence I’ve been holding briefings with James Governor since, well we argue over the exact year, I think 1995, James thinks later. However, he as much as anyone knows the value of partitioning and in fact wrote a piece on mainframe virtualization for the launch issue of the UK InformationWeek back in October 2000, I think. Then there is the Grid question. I’ve briefed Will Fellows at the 451 Group a number of times on the strategy and evolution of Grid and specifically Grid standards as they evolved and merged with Web Services standards.

Grids are a highly effective means of addressing specific business problems by splitting work up and dispatching it across a number of different systems to compute or calculate results in parallel and then bring them all back to together. Grid, in essence, is one way to virtualise a business problem. The challenges for most organisations though is deciding how to implement a grid and using what technology.

The real point is of course that all these technologies are coming together; organisations shouldn’t be forced to chose a hardware platform or build a seperate rack or cluster of systems to make up their grid. They should be able to use virtualised hardware in the form of partitions or virtual machine hosted on systems with available capacity, often underutilised, and on Systems which can acquire capacity dynamically, provision and dispatch new server instances quick and efficiently when needed. Something System p and mainframes do second to none!

The big advantage of doing this via partitioning is that you run your paralell work alongside existing production workloads without interferance, in seperate logical partitions. It can mean no additional hardware to buy, and typically you can mostly run Linux for your Grid and parallel work, it can run either clustered or isolated. AIX can also be used, and is especially interesting when it comes to exploiting existing workloads and clustering.

It’s not that mainframes would be better in a grid architecture than Sun or HP, more its about how to put your existing infrastructure to better use, and thats something we’ve been working on for a while at IBM, both in IBM mainframes and System p servers!

Funny old world the blogsphere, what goes around, comes around!