Adventures in systems land

There’s an excellent analysis by Frank Dzubeck over on Network World today about the new Enterprise Data Center and that hoary old chestnut latency. I don’t know who briefed Frank, it wasn’t me, Jeff and I talked this afternoon and I asked, it wasn’t him, since the article covered also the z10 announcement, I have a good idea though

Frank covers ensembles, data center utilization and the some of the new data center fabric issues extremely well. He also makes the point, that I’d like folks to be clear about, that this isn’t the resurgance of the mainframe, or everthing back to a central server.

We’ve grown use to indefinite waits, or unbelievably fast response times from certain popular websites, but the emerging problem is around latency in the data center. How to deliver service levels and response times in an increasingly rich and complex systems environment. It’s OK to build a data center or server subsystem focussed around a single business model, something like Amazons EC2 or S3, or Googles search and query engines; it’s another to integrate a vast array of different vendors IT equipment bought at different times for different business applications and services and integrate them all together and orchestrate them as business services. While MapReduce may or may not be as good as, or better than a database, not everything is going to be run in this fashion.

Fibre channel over ethernet is a going to happen, 10Gb ethernet opens up some real options in terms of both integrating systems, and distributing services. It will be almost as fast to connect to another server as it is to talk between cores and processors within the same server. This disclosure from IBM Research today shows the way to the next generation of interconnected infrastructure, working at 300-Gbit/second, the bus goes optical making the integration of rich data systems video, VOIP, total encryption of data, secure key based secure infrastructure services, integrated with more traditional transactional systems a real possibility.

The opportunity isn’t to take the same old stuff and distribute it because the fabric is faster, it’s about better integrating systems, exploiting new ways of doing things. Introducing a common event infrastructure, being more intelligent about WAN and Application routing, having a publish/subscribe/consume model for the infrastructure and genuinely opening it up and simplifying it.

Of course, there a re lots of blanks to be filled in, but the new Enterprise Data Center is taking shape.

A couple of things from the “Monkmaster” this morning peaked my interest and deserved a post rather than a comment. First up was James post on “your Sons IBM“. James discusses a recent theme of his around stackless stacks, and simplicity. Next-up came a tweet link on cohesiveFT and their elastic server on demand.

These are very timely, I’ve been working on a effort here in Power Systems for the past couple of months with my ATSM, Meghna Paruthi, on our appliance strategy. These are, as always with me, one layer lower than the stuff James blogs on, I deal with plumbing. It’s a theme and topic I’ll return to a few times in the coming weeks as I’m just about to wrap up the effort. We are currently looking for some Independent Software Vendors( ISVs) who already package their offerings in VMWare or Microsoft virtual appliance formats and either would like to do something similar for Power Systems, or alternatively have tried it and don’t think it would work for Power Systems.

Simple, easy to use software appliances which can be quickly and easily deployed into PowerVM Logical Partitions have a lot of promise. I’d like to have a market place of stackless, semi-or-total black box systems that can be deployed easily and quickly into a partition and use existing capacity or dynamic capacity upgrade on demand to get the equivalent of cloud computing within a Power System. Given we can already run circa 200-logical partitions on a single machine, and are planing something in the region of 4x that for the p7 based servers with PowerVM, we need to do something about the infrastructure for creating, packaging, servicing, updating and managing them.

We’ve currently got six-sorta-appliance projects in flight, one related to future datacenters, one with WebSphere XD, one with DB2, a couple around security and some ideas on entry level soft appliances.

So far it looks like OVF wrappers around the Network Installation Manager aka NIM, look like the way to go for AIX based appliances, with similar processes for i5/OS and Linux on Power appliances. However, there are a number of related issues about packaging, licensing and inter and intra appliance communication that I’m looking for some input on. So, if you are an ISV, or a startup, or even in independent contractor who is looking at how to package software for Power Systems, please feel free to post here, or email, I’d love to engage.