Anyone ever try this.....Disk I/O Performance test on DL360 G6 with P410i w/1GB FBWC

[COS]

OK, so I am always trying to tweak out as much performance out of my LAB DL360 G6 hosts as I can.

I have the hardware below...

DL360 G6

2x X5550 2.66GZ 8MB Cache

64GB PC3 10600

3 300GB SAS 6b/s in RAID 5

1 60GB SSD OCZ Vertex II SATA 2 (Max Read: 285 MB/s, Max Write: 275 MB/s)

2 Patriot Blaze 120GB in RAID 0 (555 MB/s read, 535 MB/s write) [I really didn't use these in the test]

I use SQLIO to do disk performance testing because I'm primarily a DBA.

Now, I built up the one host with just ESXi 5.5 U2 HP build.

I spun up a Win7 VM and ran the SQLIO test 3 different times and got the below results..

VM On Spindle and ESX on Spindle (This is my base lower control measurement):

IOPS: 1189, 1246, 1262 (Avg: 1232)

Transfer Rate MB/s: 74.3, 77.9, 78.9 (Avg: 77.0)

That's typical.

So, my servers do not have a CD/DVD but the motherboard has a SATA header on it. I'm guessing it's SATA 2 but if anyone knows what it really is let me know.

I figured, let's make use of it since it's there. So I bought a Slimline SATA Female to SATA Female ( 13 Pin SATA Female Slimline to 22 Pin SATA Female Cable Adapter 20 Inch | eBay  ) and plugged in my OCZ Vertex II to it.

I rebuilt my ESX host with just the SSD attached (not mounted as a datastore) and got the results below...

VM On Spindle and ESX on Spindle but attached (not mounted) SSD

IOPS: 1216, 1278, 1277 (Avg: 1257)

Transfer Rate MB/s: 76.0, 79.9, 79.8 (Avg: 78.5)

That's close enough to the first test to be no change in performance.

So next I installed ESX on the Single OCZ Vertex II SSD (that's connected to the SATA II port where the CD/DVD would normally be) but left the VM on the spindle LUN and I get the below results...

VM On Spindle and ESX on SSD

IOPS: 1338, 1365, 1369 (Avg: 1357)

Transfer Rate MB/s: 83.6, 85.3, 85.5 (Avg: 84.8)

So check out the results!  Average performance is up by a few MB/s ~6MB/s!!! :smileydevil:

Remember, the VM is still just on the Spindle LUN and NOT on the SSD.

My final test was to have both the VM and ESX on the Vertex II SSD and below is my control test for upper limit...

VM On SSD and ESX on SSD (Upper Control)

IOPS: 2580, 2592, 2585 (Avg: 2585)

Transfer Rate MB/s: 161.2, 162.0, 161.5 (Avg: 161.5)

That was as expected also because they are all on theSSD.

So in a nutshell, just putting ESX on an SSD improved my disk I/O on my spindle datastore.

Has anyone ever tried this and gotten the same results or is this an anomaly?

I did the same thing on my other DL360 G6 host and the results are the same.

Can anyone explain why my performance on the spindle datastore improved?????

[Alistar]

Hi there,

first things first - ESXi's VMkernel does not care about installation location. It gets loaded to the memory (and kept there) on the boot-time so the discrepancy between the test results must have come from somewhere else.

Did you force ESXi to identify the drive as an SSD? VMware KB:    Enabling the SSD option on SSD based disks/LUNs that are not detected as SSD by defaul...it provides the following:

■ It enables usage of SSD as swap space for improved system performance. For information about using SSD datastores to allocate space for host cache, see the vSphere Resource Management documentation.

■ It increases virtual machine consolidation ratio as SSDs can provide very high I/O throughput.

■ It supports identification of virtual SSD device by the guest operating system.

The initial "poor" performance could have been because of trimming or garbage collection running in the background - the next try might have been thanks to the "clean" disk that was present after the installation. After all OCZ recommends leaving at least 10% of the disk space free and for 60Gig SSD there could have been some recuperation time for the data to be redistributed evenly across the cells.

[COS]

ESX identified the SSD itself. It's not in a RAID set nor is the controller RAID configurable of any type because the on board SATA controller is dedicated for a CD/DVD, passthrough if you will. I did not try and have it spoofed/tricked.

I re-created the tests 2 other times with consistent results on the same server. Then I did it on another server and got the same results, faster with ESX on the SSD.

After that, I reconfigured the rest of the DL360 G6's to the same, esx on the SATA SSD where the CD/DVD should be......lol

Also, my 60GB SSD's are only SATA 2, I wonder if I get more performance if I put a SATA 3, I'm still not sure if the on board controller is SATA 2 or 3.

I'll try it on the 5th server, my guess is no performance increase but we'll see....

I'll try it on the Gen9 DL360's when they come in next month too, if they even have a SATA interface......lol