Poor disk performance with ESXi on Dell PE1950s

[dustinfl]

Hello VMWare Community. This is my first post, so bare with me. We first started off running VMWare Server 2 on a Dell PE1950 III with 32GB of RAM, 2 quad-core Xeon L5420s (2.5Ghz/core), and local storage (2 1TB drives in RAID-1) on top of Windows Server 2008. The performance was pretty good, but we wanted to try out ESXi so that we could get rid of having to worry about the underlying OS. We threw ESXi on another PE1950 I, although this one has 18GB of RAM, 2 dual-core Xeon L5148s (2.33Ghz/core), but same storage setup. However, when running tasks via SSH (I enabled it) such as trying to copy a folder in the datastore, the performance is horrible. For example, when I tried to copy a 4GB image (cp -R folder newfolder), it took roughly 90 seconds, which seems slow. Well, fast forward a couple weeks, and we just got in 3 new maxed-out PE1950 IIIs, one with 4 15k SAS drives, and the others with 2 1TB SATA drives, with the higher performance system destined to be a database server. I wanted to see what the performance would be with ESXi on the system that has the 4 SAS drives, so I created a RAID-5 container and threw ESXi on it. Again, the same exact dismal disk performance. Maybe I have high expectations for the disk performance, but it just seems a bit slow to me. The reason it concerns me is because I'd like to create a backup script that stops a virtual image, copies the contents, starts the virtual image back up, and then ships off the backup to a backup server. If it takes 45 minutes to copy a 100GB image, this is going to be a lot of downtime for some clients. We will soon have a centralized iSCSI SAN device, but for now I'd like to see if you guys think there is an issue or not. Thank you for your time.

[dustinfl]

I'm reading about the vmfs3 and the related performance issues when not talking to it properly. Ignore me until I'm well versed

[tom_e_reynolds]

Dustin,

Did you ever resolve your performance issues with the disk access on the PE1950?

We had an older PE1950 with 2x146 GB SAS 10k drives running Win2K3, with concurrent reads and writes that averaged about 80-100 per second. My tech worked with Dell to upgrade the firmware on the PERC 5 Controller, as well as upgrading the firmware on all the drives. Now we see upwards of 400 concurrent reads and writes per second, RAID1. With RAID0, we saw upwards of 810 reads and writes per second. I have 3 other PE1950s, they all average about 400 r+w per sec in RAID1.

I also know for a fact (due to 2 days of testing all drives in all possible RAID configs) that the PERC 5 controllers will max out at about 1000 reads and writes per second. I proved this with a PE1950 that had 4x75GB SAS 10K drives.

In that optimized config, I was only able to move 5.5 GB of data in 90 secs. So you ( at 4GB / 90sec) may just be at the limit of that controller and disk combination.

(my newer HP DL380 G5 will get 650 r+w per sec with 75 GB 15K SAS HP Smart Array P400/256MB Controller RAID1)

I just loaded ESXi on this PE1950 with 2x146 GB SAS 10k, I will report back if the performance changes with VMWARE loaded.