Thoughts on Virtual Infrastructure Management

Server Tiering? What the heck is that!

With it being the new year I figured I would try to impart something new.  Or maybe instead share something from the “everything that is old is new again” file.  I have found a renewed interest in storage tiering among our customers.  This is not too surprising with the folks at EMC rolling out FAST and the continued success of these in the array tiering of companies such as Compellent, HDS, and NetApp. 

This is not “repacked Information Lifecycle Management (ILM)” but instead organizations are trying to figure out how many IOPS applications are really doing, what needs Solid-State-Disk (SSD), what will be just fine on Serial ATA (SATA), and then go with Fibre Channel for the rest.  Within those tiers they are also mixing RAID types; such as RAID-5, RAID-DP and RAID-6 for bulk, RAID-10 for higher throughput read, and exploring thin-pooling / thin-provisioning to conserve more expensive disks.  But what the heck does this have to do with servers?

I am seeing more customers looking to do the same on the server side and wondering where to start.  For the most part, 90% of the organizations I talk to allow uncapped access of servers to resources.  And still far too many organizations are doing in-place virtual server provisioning.  What is in-place virtual server provisioning you might ask?  It is the craziness of taking a physical server that you are going to virtualize and provisioning in the virtual world the same number of cores and memory the server had in the physical world.  The definition of insanity is doing the same thing over again and expecting a different result. 

Weren’t you the guy who pushed back when they ordered that server without you knowing for the project?  I can hear you now saying “just because the vendor said it needs that much doesn’t mean it can really use it”.  Guess what?  You were probably right.  Why not take this opportunity to fix that problem?

This takes qualitative numbers … aka facts!  Being able to say “Mr. app owner, you have not used more than 25% of the four cores you currently own,” and “when we virtualize we will give you double your current usage,” meaning two vcpu’s worth.  Take a deep breathe … now doesn’t that feel good?

Already down the virtual-first path you say?  Well here is a perfect opportunity to go back and prove that you can get better than 10 VMs per one host.  Here is a perfect time to go and beat the industry average.  Again this is average and not all applications are the same.  So, gather the quantitative data first then make your adjustments.  And get management buy in first and I discussed in my last blog.

Once you know what the apps really use then you can tier them.  I would suggest using resources pools within VMware.  Maybe you have been using them as folders for certain types of applications.  Most people are not using them for anything more than organization.  Here are the rules of thumb I would suggest:

  1. Gather quantifiable data on CPU, Memory, and IO
  2. Keep your tiers simple – no more than three types / tiers
  3. Create specific resource pools for high-end mission critical, “I-lose-my-job-if-there-is-an-issue” applications that have minimum service levels
  4. Create a resource pool for everyone-else (the middle-class) and maybe set no upper or lower reserves or caps
  5. Now create ones for those servers that are allowed to be “slow” and put caps on them
  6. Measure your success (see #1 and compare) – you should be able to provide service levels to keep even the most vocal critic happy

As always, please share ways you have been able to bring service levels into you virtual environment.  Also, let me know how you define server tiers in your organization.

January 11, 2010 | Filed Under Virtual infrastructure management | Leave a Comment 

comments

Leave a Reply