Monday, 13 April 2009
Virtualisation of Production Workload Hurdling - Olympic Sport?
This isn't a jibe or criticism at a vendor (for once), its not a post to single out any one company or individual, its a post to see what possibility may lie ahead for anyone who is in the early stages of Virtual maturity and are just starting to embark on a virtual journey or even go from Virtualisation being a point solution and into more of a fundamental enabler of your infrastructure ecosystem. It will provide you with some pointers and tips on what to do when faced with wanting to push the boundary of virtualisation into Production.
The Wow factor
Ok so you've virtualised 50-60% of your estate, mainly most of this is likely to be Dev and Test, and your cooking on gas, your finding it easy to deploy new VM's for your organisation needs, you can ensure you have clarity on when you are due to run out of capacity due to extensive monitoring capability, you can snapshot and make up for development team errors and cockups, backups are a breeze with VCB, DR for you would seem so much easier when used on your complete estate, and many more in built benefits of the core tech.
Importantly the Infrastructure team are the flavour of the month, your developers actually now talk to you at the coffee machine rather than grunt when they pass you in the corridor, they are your new best friend as hey you can spin them up a VM in hours not days you can make up for their mistakes with snapshot reverts, you can make up for there poor planning etc etc.
Your senior members of staff however see this new found ease and benefit differently, and most likely dont even care about the technical benefits, they maybe have provided the funds you to virtualise in the first place.
So after the success of the Development environment consolidation, the next natural stages are you are looking to virtualise your Prod workloads and need to go back with a solid case to Virtualise and increase overall Physical Infrastructure to support the consolidated environment. This is where it starts to become a problem, The P word now means core solid investment being needed, something any company will struggle to do if the current physical environment is stable, solid and able to facilitate the business need. Doubts and hurdles you may find are as follows;
"But the current hardware is a year old and needs to be depreciated"
This will be hard for anybody to argue against the kit being a sensible candidate for consolidating, would you sell your current petrol car and go and buy a diesel just because it paid back over a period of months and was easier to run, maintain etc. Chances are in that analogy no (and specially in todays car market), however possible arguements you may find benefit the decision in the server world include;
- Reductions in Network and Storage connectivity
- Removal of any breakfix maintenance on that box
- Ability to now provide High Availability and zero downtime with VMotion/HA/DRS on a platform that was unable to have this in the physical world due to cost etc
- Ability to clone into a Development environment in pristine fashion, not by a lenghty backup and restore process etc
"My platform isnt suitable for Virtualisation due to performance"
As you may have experienced with any introduction of new technology this will be a common one, and hey as your the new guy with the new ideas it will be you that has the "its his fault for bad app performance" tagged to your Active Directory profile for all of the company to see that its your fault, hey if your gona virtualise production you gota be used to putting your head on the block remember :).
With recent developments in new generation server platforms which are now tailored for Virtualisation such as HP Proliant and Bladesystem G6 Ranges this issue is now becoming less of a problem for people to consolidate more IO intensive workloads. Couple this with improvements which will come in VMware Vsphere VI4 and it will no longer be as much of a problem (I didnt say it wont be a problem). Remember with Consolidation you are still slicing and dicing a physical host, this has a layer of indirection which will if not suitably planned cause you problems when the workload demands more compute power.
Companies adopting virtualisation really should invest in tools which allow you to analyse the complete estate, VMware capacity planner can do this, Novell Platespin Power recon is also a good tool to use, however some companies cannot afford to do this and larger companies may do a scoping exercise 6-12 months before they actually start to consolidate workloads so it cannot cover all workloads, within this time you may have had new applications put in, or your business requirements and processes may have changed which means your application workload is increased.The key to this are obvious ones but maybe of use, these include;
- Get to know your application, its no good just asking the app bods for the thumbs up, they most likely do not know the infrastructure layer components and the four main bottlenecks that affect virtualisation performance, Network, Disk, CPU and RAM. Using monitoring tools before this in even the simpliest form such as Perfmon will enable you to observe the platform in more detail and observe factors such as Page Faults, heavey disk paging, high Disk queues and excessive process threads.
- Test test test, yes you can do this. Building a Virtual Machine and OS from scratch and then asking application teams to install the app on a clean build is an easy reality for you to be able to test and derisk. If your looking for a quick turnaround and are confident that you only need a slight amount of confidence instilled into application teams, then one option is if you have dev and test environments then using a P2V tool such as VMware converter or Platespin Powerconvert to migrate your physical workload to test in a quarantined environment will enable you and your application bod to ensure you know how the workload will perform and its possible problem areas when virtualised. The alternative to testing in a dev environment is to completely reinstall the application from scratch, this would be my prefferred.
- Consult the ISV, these guys will if a big software house be able to provide at least some partial advice on performance profile which is expected, take some of this with a pinch of salt though. Unless they are a SAP or an Oracle mould then chances are they will not know Virtualisation as indepth as you, hey didnt they get us in this mess in the first place by speccing highly rediculous minimum hardware specifications :)
Other core drivers
Other drivers to virtualise production workloads which have been implemented recently maybe;
- Turnkey DR, DR may not have been in scope of the project so was not funded, remember you can easily replicate VM's now, something that provides 1-2 Hour RTO Times against what is probably if no funding was provided a 24-48 Hour RTO from backup process.
- Snapshots and instant restores for Future software upgrades,
- Improved Performance monitoring being available, improvements in future releases with Vmware Appspeed will enhance this even further.
Well thats enough hopefully to kick you off, obviously being the VMware evangelist you are now and knowing how much confidence you have in the product the above will assist, your passion and drive will hopefully instill confidence within itself. You know the more extended benefits brought to the table in areas of Business continuity and agility, something which is usually a key driver.
Subscribe to Posts [Atom]