1

I’m building a configuration for a Dell R720, to do some bench marking of our software with a larger SQL data set than i’m used to, we’re planning to test from 200GB up to 1TB.

My budget can’t stretch to the SAS SSD’s or Fusion IO cards so i’m left with raid 10 for sata SSD or 15k spinning rust over 6Gbps SAS.

Has anybody got an idea of how the SATA ssd’s compare to the 15k 6Gbps SAS drives?

Am I going to notice the difference? Is it worth it for the loss in capacity?

@Dell_Technologies

7 Spice ups
2

I guess it would depend on RAID levels etc.

For example, if this is only for database storage, you could just put in a few SSD’s

Or you could try to reduce costs by putting File systems/Operating systems etc on slower SAS disks in one RAID set and smaller SSDs on another SAID set.

From memory you can get 16x 2.5 inch disks in a 720, leaving room for later expansion.

3

I’m going to venture some thoughts from what I know and read, but as to actual testing between the two choices, I haven’t seen any real world comparison. First “It depends” is the universal answer, we all know any particular build is special. However, some things can be deduced.

First, your controller can be a limiting factor, regardless of drive’s capability. Some RAIDed drives can be slower tahn single drives. But say you have a ‘perfect’ controller with enough HP and channels so as to not limit anything. Between drives, the theoretical thruput per SAS or SATA channel will be approx. the same. However, a ‘good’ SSD can approach that channel limit for thruput, i.e., 500+MB/s is not that uncommon. A good SAS Seagate Cheeta 15.7K drive is spec’d for a max continuous 204MB/s. So no contest, right? But the caveat is can the SSD keep its thruput up? The HD drive can, but SSDs are different. There are enterprise SSDs that are made for that type of abuse, but many are not. So bottom line, which SSD (as well as which HD) you pick can produce wildly different results. IF your controller can take it all and deliver it and/or If your data demand is intermittent, the SSD with orders of magnitude better response time, and a few times better thruput, should win, in almost any way you can test. So - no consumer drives, only enterprise SATA SSDs, or better a nice dedicated PCIe SSD Card. Anything to bypass SATA helps. Read about one new entry here >> www.micron.com for a SysBench test of a 264GB database with 100 tables and 1 billion rows, a 24GB buffer pool, and 32 threads. Impressive and what well be seeing more of in the months and years ahead!!!

1 Spice up
4

If you need random access, low latency and IOPS go flash (database and VM-centric workload). If you need linear read and write performance go RAID-ed spindles (video capture, backup and replication).

1 Spice up
5

@Carl Potter,

Here is a Dell techcenter study on optimizing server/ storage performance on R720. I think it has IOPS charts that you are looking for: Computers, Monitors & Technology Solutions | Dell USA

Hope this helps!

Thank you,

Ivan@Dell

1 Spice up
6

What’s the budget if you don’t mind me asking?

I’ve seen some pretty good pricing on 6Gb/s SSD’s which I have purchased myself and put in my own machines for RAID. They were not that much, do you need over a TB of storage it seems, perhaps 1.5-2.0TB since you are doing testing up to 1TB and need headroom for the OS etc. ?

Also, unless you have a specific reason, yes the FusionIO cards are very expensive but there are many other options out there I use with several clients, including myself that get the job done very well.

Best,

Chris Gentile

Expert Server Group

603-714-3274

cgentile@expertserver.com

7

I’ve managed to get the dev team to scale back the data size, so i’m looking at 4 * “Dell 100GB, SSD SATA Value MLC 3Gbps” which come in at approx £310 each.

Because it’s not in production i’m planning to run them in raid 0.

8

For a lab/benchmark rig, just use off-the-shelf SAS MLC SSDs.

1 Spice up
9

RAID0 on SSD is pretty safe anyhow…