So, you’ve discovered server virtualisation and think you’ve found the answer to all your infrastructure problems; Couple of HP Servers, VMware and point it at some disks.. wrong !!
All to often I come across customers who have virtualised their server estate, then pointed their servers at an MSA1000 full of SATA.. then for some unknown reason exchange runs like a dog. databases timeout and end users get an all together bad experience of Virtualisation which leaves a bad taste in the mouth. Why does this happen ?
In the pre-virtualisation era, Mr.Blogs had 15 servers, each with 10k/15k SAS disk.. maybe 30 disks or more across the estate. These disks have sufficient performance capability to handle the disk load each server and its respective applications puts on its internal disk. Now Mr.Blogs decides that his server estate will be better utlised, easier to manage, will consume less power if he virtualises. All very real benefits of virtualisation… but Mr.Blogs realises that he won’t be able to capitalise on many features like VMWare’s VMotion, HA and lockstep without consolidated storage. So, he goes on a hunt for some cheap storage, budget and capacity his focus, he’s led to believe that SATA having the best price interms of cost per GB is the way forward. So, out comes the refurbished MSA1000 and all his VMWare Datastores go on one big RAID group.. job done ! no…
I apologise if I’m telling people how to suck eggs here.. but I’m going to serve you up a nice big hard boiled one. Applications in virtual machines have the same Disk IO requirements as Applications on physical machines and SATA is typically not a good substitute for SAS or FC disk that once serviced these applications in the flavour of local disk. The same best practices must be observed (seperate Random and sequential type data onto seperate spindles , blah blah blah, etc etc), exchange will still hammer drives in the same way it allways has and backup jobs will still consume disk bandwidth.. so what do we do ?! fast spinny disk is expensive !!
Gather data relating to your applications requirements, size accoringly and identify the actual disk requirements required for performance not just capacity (talk to a storage reseller or vendor if you need help with this), if possible tier your storage, implemement some level of information lifecycle management, apply some quality of service.. just be aware that Mr.Blogs approach will result in nothing but disk contention and large application response times.
If you want to look at sizing accordingly, view my post on Sizing for Performance. Some data wil happily sit on SATA, test and dev stuff, file system data, VMWare ISO templates, archive data, B2D data.. but I would always recommend having transactional database data on SAS or FC (or now SSD, if you got the money.. but thats a whole new topic).
In terms of managing your data’s lifecycle; there are many products out there which will move your data to a cheaper tier of disk by policy, maybe start with mail system archive; stub and archive mails over 90 days old to a cheaper tier of disk (keeping capacity requirements for fast spinny disk down and potentially eliminating the requirement for managing mailbox quotas and PST’s !! 2 birds 1 stone !!). Look at File system data (I can almost garauntee that at least 60% of it has been inactive for 6 months or more). Archiving will allow you to keep data online and accessable to users, just means that archived data might take a few seconds longer to get to their screen. Another benefit of this is that most archiving products will index data as its ingested meaning that there are benefits around discovery also.
Also, much like VMWare’s DRS (distributed resource scheduler), many storage products have the ability to apply QOS at the disk subsystem layer. EMC have QOS Manager for Clariion, which will talk to DRS and will give you the ability to set a Goal for a specific application, so that it allways has a certain amount of disk bandwidth. Alternatively it can be used to ensure that backup jobs don’t hammer the disks so hard that your applications fall over..
In any case, I would recommend that rather than just jumping online and buying storage from a company that shifts tin in a similar fashion to ready meals from sainsburys; take a look at the al a’carte menu and do it properly, you will save yourself a post sales nightmare.
It may cost a little more in the first run…. but a cheap solution is only cheap until it fails.. then it can become very very expensive.