Whether to use SAN or local storage for best High Availability for Esxi environment


I should probably lead into this by saying that this is my first production virtual systems rollout, so I appreciate your feedback and patience.  I am configuring a virtual environment for a client that is currently running all physical hosts.  It is a small firm, c. 15 users total, but the owner is very tech-oriented and forward thinking.  Currently they are working off the following physical hosts:

1) Exchange 2003 server + DC
2) SQL Std 2005 server + file server
3) anti-spam server SMTP gateway
4) 2X server (similar to Citrix)

I am looking to condense these into 2 Esxi hosts that will be running the above server functions in an HA environment, along with an APC VMA server on each host that will generate an auto-shutdown in the event of power failure.  So there will be 8 VM's in all.  I plan to have 4 VM's operational on one host and 4 on the other host to split the regular workload, and in case of a host failure all running VM's (except the APC) will migrate to the remaining host.  We'll be using either FT or HA to ensure either zero or minimal loss of uptime for all VM's across the hosts.

My question is one of architecture and reliability.  One option I'm looking at is having 2 hosts with 96 GB of RAM and Esxi on SD media, and a Dell MD3220 SAN for shared storage, housing all VM's, data drives, etc.  The second option is to equip both hosts with local storage, and setup the same failover scheme across both hosts.  With the current configuration, price is c. 52K for the SAN + diskless hosts, and c. 44K for the 2 hosts w/local storage.

For this type of setup, is there a prevailing recommendation across industry pros on whether to SAN or not to SAN?  I've read in a number of tech forums that some storage veterans recommend local storage vs. a SAN since it's a single point of failure and adds too much to the budget,  Definitely no small difference in price as per above.  However I've also read accounts of SAN's that run for years without ever power cycling them.

Many thanks for your opinions and thoughts,

Who is Participating?
Paul SolovyovskyConnect With a Mentor Senior IT AdvisorCommented:
We've had some performance issues with the VSA, make sure you test first and gather IOPs/latency performance.  You may also lose over 50% of your storage if not more

(RAID 10 - lose 50% on each host) (Replication - lose that volume on the destination host  for each volume replicated )..so you may wind up with 25% of total capacity..
chlebiTeam Leader IT InfrastructureCommented:
I don't quite get how you would maintain data on local storage if one host fails.
The point is that VMs running on one host would write data to that particular host. Then if such host fails (so its local storage is not available) you won't be able to start the VMs on the other host as it does not have the actual data. For HA you definitely need a shared storage. Or am I missing something?
jkirmanPrincipalAuthor Commented:
Hello  chlebi,

Sorry I didn't clarify, the local storage option I mentioned is VSA.  It looks like it will turn the local storage of hosts into a pooled resource, and basically turns pools of local drives across 2 or 3 hosts into the functional equivalent of a SAN.  It will mirror data across hosts.  HA, FT, Vmotion, DRS, and Storage Vmotion are supported, depending on the version of vSphere.
jkirmanPrincipalAuthor Commented:

I checked with my Vmware rep and his SE, and the SE recommends against going with VSA.  In his words, it is "not.. awesome", plus VMWare will stop selling it in about a month.  It will be supported for 3 more years, but apparently it hasn't taken the SMB market by storm.  I'm guessing that using a software mirroring platform between servers generates a lot of overhead, and he said he particularly wouldn't recommend it for a SQL application.  When doing the price workup for a SAN approach vs. the VSA, and considering I'd have to put in 2X to 4X the amount of storage (RAID-5 vs. RAID-10) that I need because of how VSA works, the price difference in the end was not that much, so SAN remains my primary recommendation to the client.  FWIW I came across conflicting specs on which disk configuration VSA supports, since a VMWare doc show various supported RAID-5 and RAID-10 configurations, but some docs say RAID-10 is required.  For an alternate approach, I'll be proposing a single active ESXi host running all VM's on local storage, and a less beefy but still capable ESXi warm spare where the VM's are kept current using Shadowprotect's HSR Head Start Restore product.  Not an HA solution, of course, as it will require manual startup of the warm VM's, but it does in the end protect against host failure.  I may even suggest the HSR option for the SAN configuration to protect against the very unlikely possibility of a SAN failure, although that may totally break the bank.  Shadowprotect's ImageManager already includes Oracle VirtualBox to allow booting a VM off the SP backup images so is already a minimum but available Business Continuity solution.

Awarding points,

Paul SolovyovskySenior IT AdvisorCommented:
All depends on RPO and RTO.  Some customers of ours use replication as a cheaper way to do business continuity while others use SANs such as Netapp, Nimble, HP, etc.. that allows shared storage, even though the SAN is typically a single point of failure it has multiple redundancies to allow it to withstand an outage of most components without losing data
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.