Anyone using SSD's in Raid for VM arrays ?

V

VDiesel

Weaksauce
Joined
Apr 16, 2008
Messages
94
Anyone using SSD's in raid 5 for VM storage ?

I'm currently using Windows Server 2008 R2 with Hyper-V running 5-6 vms in a SOHO environment on 4 x SATA 640GB WD6400AAKS drives in raid 5 with an Areca 1261ML. While it runs ok, I can't help but feel that drive contention is an issue at time, which is inherent to the way SATA works.

I could of course change raid HBAs to a SAS controller and pick up 4 x15k SAS drives, but I could also keep the Areca if I went SSDs and of course benefit from their great random read/writes.

For the same money I was thinking of getting 4 x Intel 160GB X25-M G2's in R5, but my main concern is I don't know of any raid HBA that pass TRIM/GC to the SSD in an array. Will this cause things to degrade over time. I'd hate to spend that kind of cash and end up with slower than single disk speeds in a few months.

Thoughts ?
 
Last edited:
I have been trying (unsuccessfully, for now) to convince my boss to allow me to get some small OCZ Vertex SSDs for my development Laptop (Sager NP9262, currently with 3x 5400RPM 160GB drives) to assist when I install virtualization platforms for testing. By the time I get the management VMs up and running, or if I am doing Type 2 stuff, the drive contention limits me from doing a lot of things. RAM is also an issue, but that's another thread, I suppose. I am hoping that with end of the year money we can do an i7 laptop loaded with RAM, and 3x small but fast SSDs as my mobile dev platform for virtualization.

with regard to your post, I am not sure what to tell you. I think I would go with the sure bet, for the same money, and go SAS with 15k drives...but it also depends on your usage. Are you noticing perceived contention when performing a specific operation, or only on certain VMs? is it possible to move the less active VMs to the same spindle, giving the more demanding ones their own? Are your currently using a RAID level on the 640GB drives (you didn't mention one)? If so, I would recommend either adding more 640GB drives (spindle count FTW with SATA) to the array, or breaking the array ad going with 2x VMs per spindle as a rule. You might also consider breaking the array, ditching the 320GB drives for 74GB raptors, and running 2x per spindle (or 150GB, or 300GB raptors) depending on your size needs. However, that would only stand if you are doing this at home.

If this is a work production environment, then I would certainly not make a recommendation to use SATA drives at all, unless they are enterprise class drives, and in a SAN, really. if this is a production environment, then I would definitely go with a fully supported solution, checking the HCL on all components.

there's a lot of variables to be considered, but we can't answer the question without some more detail, really.
 
sorry updated my original post with some info.
It's a soho vm setup - ie not enterprise.
The WD640's are in raid 5. The Areca 1261 has a 4GB Dimm ECC cache.

For all the reading I've done 4 x Intel SSD rip apart 4 x 15k SAS drives in random reads and writes. Then there's the considerably lower power consumption of SSDs and higher reliability of solid state vs mechanical spindles.

I'm just wondering if anyone has setup SSD's in raid 5 for VM storage and how the experience has been. Would the lack of TRIM/GC be of concern in this setup ?
 
lopoetve said:
EMC just lauched a new one. 180,000 IOPS using SSD :)
Click to expand...

Well actually it's not just EMC, they're just rebranding it into their solution.
HP, Sun, Hitachi and others are all joining along for the ride.

It's from a company called Pliant that just unleashed some serious enterprise flash drives (EFD)s that can produce 180,000 IOPS with 500MB/sec sustained read or 320MB/sec write rates each.

But of course these drives will be priced even higher than the Intel SLC X-25E's, well north of $800 for a mere 64GB.

http://news.techworld.com/storage/3201740/solid-state-drives-offered-for-emc-hp-sun-hitachi/
 
I'm actually thinking I might keep the 4 x 640gb raid 5 array on the areca for less I/O intensive VMs and just add a single Intel X25M 160GB to the ICH to run 1 or 2 more I/O intensive desktop VMs and SQL.

That would most likely be the most cost efficient way of doing it and still manage to keep TRIM support on the SSD (when it's released).

I just don't know how I feel about running critical VMs on a single drive, without fault tolerance for drive failures.
 
Ditch RAID5, that will eat up your writes on SSDs. Probaly will be cheaper to get 4/6 MLC SSDs in RAID10 rather than 3-4 SLC SSDs in RAID5. But whatever.

I recently (whenever Vertex drivers first were released) switched my VMs to RAID0 Vertex drives. I will never use VMs on anything but SSDs again. Raid0 probably isn't that big a deal but SSDs just in general are awesome for VMs. I used to try to balance vms between multiple spindle disks and blah blah, how I just throw them all on a single raid0 array and everythign is just speedy.
 
VDiesel said:
Well actually it's not just EMC, they're just rebranding it into their solution.
HP, Sun, Hitachi and others are all joining along for the ride.

It's from a company called Pliant that just unleashed some serious enterprise flash drives (EFD)s that can produce 180,000 IOPS with 500MB/sec sustained read or 320MB/sec write rates each.

But of course these drives will be priced even higher than the Intel SLC X-25E's, well north of $800 for a mere 64GB.

http://news.techworld.com/storage/3201740/solid-state-drives-offered-for-emc-hp-sun-hitachi/
Click to expand...

The only ones that will get close to that will be the DMX series I expect :)
 
Yes they will not automatically pass the TRIM command when used in a RAID environment.
Although if you get Indillix based SSDs like the the Vertex you can setup the the GC utility to run as a scheduled tasks, every night.
 
Nitro....how does this happen ?
When you setup these SSDs on raid controller the OS never actually sees the SSDs themselves, only the array (as a disk). So can software be scheduled to run GC on drives the OS can't access directly?

Also aren't the Vertex poor with random reads/writes ?
 
VDiesel said:
Nitro....how does this happen ?
When you setup these SSDs on raid controller the OS never actually sees the SSDs themselves, only the array (as a disk). So can software be scheduled to run GC on drives the OS can't access directly?

Also aren't the Vertex poor with random reads/writes ?
Click to expand...

Yea there is a GC tool you can download from OCZ, and you can schedule it to run like any other program.

Poor random IO?
They are not as good as Intel G2's, but are lightyears better than any HDD.
Also if you put Intels in RAID they wont have TRIM/GC so then the OCZs will be do better.
 
Wow as they say ...I've found my thrill !
http://www.techreport.com/discussions.x/17600

OCZ: Our PCIe solid-state drives are available
z-drive_il.jpg



2 of these in a raid 1 mirror would be awesome for VM storage once they hit the $800 range.
 
You must log in or register to reply here.
Back
Top