VMware Cloud Community
joshrock11
Contributor
Contributor

ESXi Performance Question

I have a HP DL360G5 with a single quad core 2.5ghz proc, 4 147GB 10k drives (RAID5) and 4GB of RAM. I was wondering what I should expect to see disk speed wise from ESXi on this configuration. I had 3.5 U4 on this same box and didn't think it was running up to full potential but didn't know what an average should be. With 4 on the box I see a max of about 15000 KBps when starting all the virtual machines at once but that doesn't seem very fast .

Reply
0 Kudos
11 Replies
Troy_Clavell
Immortal
Immortal

I can't answer specifically, but the 64bit hypervisor for vSphere4, you should see better performance all around.

Reply
0 Kudos
joshrock11
Contributor
Contributor

Does anyone have a good idea of how to performance test the hypervisor? I've ran performance tests on the virtual hosts but they all come back with awful results. most of them show the disk as slower than SCSI320.

Reply
0 Kudos
Troy_Clavell
Immortal
Immortal

have a look at VMware VMmark

Reply
0 Kudos
AntonVZhbankov
Immortal
Immortal

By default all VMDKs are created in "thick" mode. It means that all the space is preallocated, but not wiped, so on first access to each block on vmdk ESX wipes block first. As a result you see degraded performance for some time.

If you really need performance right now - run some program to access all the HDD first, so there won't be more wipes. Or create "eagerzeroedthick" disks by using vmkfstools command.


---

VMware vExpert '2009

http://blog.vadmin.ru

EMCCAe, HPE ASE, MCITP: SA+VA, VCP 3/4/5, VMware vExpert XO (14 stars)
VMUG Russia Leader
http://t.me/beerpanda
Reply
0 Kudos
joshrock11
Contributor
Contributor

Does anyone have any stats they can provide me from their 380s or 360s for compareables on just the disk? I only ask because the server in question has already had the drive backplane and the RAID card replaced by HP and still doesn't perform as well as our old Dell PowerEdge 2850s running ESXi 4

Reply
0 Kudos
Sanjana
Hot Shot
Hot Shot

Josh,

This may be a shot in the dark, and you may have already looked into this, but does the controller have an onboard write-back cache that is disabled, perhaps ?

--sanjana

Reply
0 Kudos
joshrock11
Contributor
Contributor

We added a battery to the system a few months back to enable the write back for the RAID controller and it made a difference but still have this issue. I changed the HW prefetch and adjacent memory prefetch along with disabling serial ports in the BIOS and it seems to be a bit faster now. Any ideas if these should remain disabled because the HP bios didn't seem to think it was a good idea but I see in benchmark testing users disabled these.

Reply
0 Kudos
Scissor
Virtuoso
Virtuoso

Which disk controller card do you have installed? I have a HP DL360G5 with a Smart Array P400i (64MB + Battery Backup) w/6 x 147GB SAS set up as RAID0. On a quick test using 400MB file I see write performance of around 25,000 and read performance around 23,000. Since you're running RAID5 you'll lose some disk speed to the parity calculations.

Rather than starting all your VMs at once, try starting them one after the other to reduce disk contention.

Reply
0 Kudos
joshrock11
Contributor
Contributor

Ok, I just start them at the same time to try and run the disks up only to check the speeds. I also have the 400 card in my box. I will probably pull all the hosts off that server and change it to RAID 1+0 or something faster than 5

Reply
0 Kudos
Scissor
Virtuoso
Virtuoso

Found this web page from a year ago reporting disk IO speeds using our P400i controller under different Virtualization platforms.

http://vmfaq.com/entry/33/

Reply
0 Kudos
Scissor
Virtuoso
Virtuoso

Found this nugget in the ESXi4 Release notes at :

Slow performance during virtual machine power-On or disk I/O on ESX/ESXi on the HP G6 Platform with P410i or P410 Smart Array Controller
Some of these hosts might show slow performance during virtual machine power on or while generating disk I/O. The major symptom is degraded I/O performance, causing large numbers of error messages similar to the following to be logged to /var/log/messages.
Mar 25 17:39:25 vmkernel: 0:00:08:47.438 cpu1:4097)scsi_cmd_alloc returned NULL!
Mar 25 17:39:25 vmkernel: 0:00:08:47.438 cpu1:4097)scsi_cmd_alloc returned NULL!
Mar 25 17:39:26 vmkernel: 0:00:08:47.632 cpu1:4097)NMP: nmp_CompleteCommandForPath: Command 0x28 (0x410005060600) to NMP device
"naa.600508b1001030304643453441300100" failed on physical path "vmhba0:C0:T0:L1" H:0x1 D:0x0 P:0x0 Possible sense data: 0x
Mar 25 17:39:26 0 0x0 0x0.
Mar 25 17:39:26 vmkernel: 0:00:08:47.632 cpu1:4097)WARNING: NMP: nmp_DeviceRetryCommand: Device
"naa.600508b1001030304643453441300100": awaiting fast path state update for failoverwith I/O blocked. No prior reservation
exists on the device.
Mar 25 17:39:26 vmkernel: 0:00:08:47.632 cpu1:4097)NMP: nmp_CompleteCommandForPath: Command 0x28 (0x410005060700) to NMP device
"naa.600508b1001030304643453441300100" failed on physical path "vmhba0:C0:T0:L1" H:0x1 D:0x0 P:0x0 Possible sense data: 0x
Mar 25 17:39:26 0 0x0 0x0.
Workaround: Install the HP 256MB P-series Cache Upgrade module: .

Reply
0 Kudos