[Webinar] Streamline your web hosting managementRegister Today

  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 458
  • Last Modified:

SAN disks needed for VMware

We are currently looking into a SAN to store our virtual machines on.  The SAN will also be used for file storage and data backup.  My question is concerning the type of disks that would be needed for the environment.  I want to put together a SAN that is both budget friendly as well as performance oriented if that is possible.  Do the vm's run on the ESX server or do they run off the disks they are stored on?  I guess what I am trying to ask is if I stored teh vm's on slower SATA disks would it be noticeable to the suer accessing that vm?  The vm's currently reside on a Dell server where ESXi is installed.  My plan would be to migrate teh vm's to the SAN and cluster the 3 ESXi servers and access vm's from SAN.  Also, what RAID is preffered for both vm storage and file storage? File storage is mostly reading.
1 Solution
Buy a controller that is on the HCL, then choose the [enterprise/server class] disks certified by the controller that meet your capacity and performance requirements.

One does NOT just choose the disks first any more than you choose the tires before selecting the make/model of car you want to buy.
Andrej PirmanCommented:
For budget friendly SAN I would recommend NetAPP FAS2220 with 12 disks 600 GB 15 SAS, redundant dual controllers and upgrade to 10 Gbit Ethernet. Despite it is not documented on NEtAPP web site, this upgrade is possible even for entry-level FAS 2220.
We sell such storage for approx. 10k EUR + 900 EUR upgrade to 10 Gbit Ethernet

You get 4,4 TB of usable storage in such a configuration, and it is more than suitable for running VM's. Even more - due to extreme good deduplication, NEtAPP might store on this config maybe over 40 TB of data, depending on type of data.
And yes, it is very fast, faster than most of other SAN's on market in this price range, and 10 Gbit option is much faster and better performing, more I/O than 8 Gbit Fibre.

VM's run of from LUN, whose performance in WRITE mode relies mostly on RAID controller Cache, while in READ mode it depends on situation. General rule is, more (disk) spindles, faster it gets on reading.
Above config has 10 disks for reading, and 12 GB write/read mixed cache, which is excellent.

From my experience, VM's on slower SATA disks are acceptable only on RAID 10 array with at least of 512 MB cache on RAID controller. This is true for SMB segment, where there are not many IOPS, for example, NOT for some heavily loaded *SQL server. I hit the bottom with 6 disks in RAID 10 with MySQL server, having approx 100 SQL queries per second.
I would also look at HP's P4300 (formally Left Hand) . Once nice thing about the P4300 is the power of it increases as you add more nodes not the other way around.

By the way NetAPP makes great products nothing bad to say about them.
Receive 1:1 tech help

Solve your biggest tech problems alongside global tech experts with 1:1 help.

Paul SolovyovskySenior IT AdvisorCommented:
I agreee with Lasby on the Netapp but 40TB is stretching it (and this is coming from a Netapp guy).  I would also take a look a the EMCe VNXe models as well, their snpashots aren't as good as the Netapp and deduplication is useless but overall a good unit for VMware since it integrates directly with the hosts and vCenter.
As a customer already using Dell equipment, you should consider the Dell/Equallogic line as well as putting NexentaStor on a pair of Dell Servers. The advantage of the Nexenta option is that you get both file-based (NFS, SMB) storage as well as block (iSCSI, FC), and it can readily scale using commodity servers and disk trays.

The speed of the disk and VM performance is related, but only for disk-intensive operations. Boot times, on-guest file access, etc, are all impacted, but its a holistic thing: if the array is backd by SATA or near-line SAS, but fronted by a sufficiently-sized RAM or SSD cache, the performance of the VMs may actually be improved over the performance you see today on your direct-attached storage. It's only when cache is "overwhelmed" that the underlying disk plays a role. Connectivity (1Gbps or 10Gbps Ethermet, 4Gbps or 8Gbps fibre channel) will also have an impact on the throughput to the array, and thus the perceived performance of the VMs.

RAID levels are a continuing source of discussion and argument in the industry. The tried-and-true RAID5 is coming under fire with 1TB and larger disks, because rebuild algorithms--not to mention failure recognition--are starting to creep into the statistical areas which will almost guarantee "bit rot" in your data. RAID6 use is growing, but the significant impact of "write penalty" means that you'll want a larger cache in front of that storage than any other. Finally, if you can, RAID10 offers the best performance, albeit at the highest usable capacity penalty.

For a datastore with a high read rate (>70%), I wouldn't think twice about using RAID6 in that scenario; if the individual disk capacity is 600GB or less, RAID5 will give a better write performance with less overhead for parity storage.
NytroZAuthor Commented:
These are all very informing answers.  Id like to share a littel more information on my environment and hopefully get some more direct answers.  We are a very small business on the inside.  There are about 20 users with about 6 developers that each run several vm's.  Nono of the users are considered much of a power user.  We only use SQL for a small Sharepoint server and a few other development projects.  We have a small Exhange 2010 with about 40 mailboxes and 30Gb db.  teh developers currently run VM workstation at desktop but we would like to move this all to the ESX servers and SAN.  Instead of buying new workstations we wold like to invest in the server/SAN.  With thsi said, what would the setup look like if we needed about 8TB?
Take a look at http://www.nexenta.com/corp/solutions/dell-and-nexenta-storage-solutions for some Dell-based Nexenta solutions.
With only 8TB and a small number of machines, i also recommend just getting a decent server class machine with enterprise class SATA disks, and Nexenta.  Configure as a RAIDZ2 (similar to RAID6).  But then also use 3 SSDs in the pool(s).  One will be a mirrored pair as a ZIL, the other as a cache.

Such will remove the performance penalty of RAID6 and let you benefit from things like hot snapshot, online compression, de-duplication.  In fact, you may end up needing much less than 8TB by the time it is all done due to dedup alone.

Featured Post

The 14th Annual Expert Award Winners

The results are in! Meet the top members of our 2017 Expert Awards. Congratulations to all who qualified!

Tackle projects and never again get stuck behind a technical roadblock.
Join Now