Bladewatch.com

Ask any of your IT management teams and they will admit (all be it possibly in a defensive tone) that the server estate at any point in time is 90% or so idle.  That means that as a result of limitations with the operating system, application code, and due to the windows where your applications are in use (like 9-5 weekdays) your servers are sitting there running the system idle process. To a business sponsor, that can mean I’m paying for something and not seeing the benefit.

This is where virtualization and (or) grid comes into place. 

With vmware you’d take those servers running at low utilization and virtualize them onto a vmware session, allocating the required number of processors and memory, putting them on a vmware box with similar operating systems, and configurations to share the processing and reclaim/reuse/recycle the physical asset. 

With grid you could decide ok, all servers join the ‘generic’ batch grid which any business line can use.  A rule is established to scavenge processing power from any server below 45% utilization, the server notifies the grid, grid submits a batch and the server does some grid processing.

We know vmware can do it, we know grid can do it, it’s a matter of agreeing a window, whether this requires more groups for different windows or better configuration of the ESX server, it’s still an interesting test I’d like to see tried.

My final ramble on the subject is about blade farms. I’ll give you a case study example. 

We manage a blade farm comprising several thousand batch calculation servers, they’re busy, they run at 100% for about 8/12hrs a day, and they are running windows 2003 with datasynapse.

They’re dual processor dual core blades with 8GB ram. I’d love to try the following on two blades:
Install ESX on each blade
Install two windows 2003 virtual machines on each blade and install datasynapse

*  I have the same amount of processors, but is this a more efficient form of providing processor engines? 
    2000 versions of the system idle process and explorer can’t be efficient
*  I would have in effect one copy of the common windows processes
*  I wouldn’t have the layered software drivers running etc. 

Something to think about, but I certainly wonder if in the near future someone thinks about instead of deploying 2000 blades with windows, 2000 ESX servers and 8000 calculation virtual machines.  (ESX could even have datasynapse installed on it?)

License wise maybe more, but if the performance/throughput is better, this might not be a problem.

No related posts.

Related posts brought to you by Yet Another Related Posts Plugin.