I have an old(er) HP DL385G1 server with a Smart Array P600 storage controller in it that I want to use for second-tier VMWare server. I easily installed ESXi 4.1_U1 on it yesterday and the install was smooth and flawless. Its not on the HCL, but a number of folks have reported installing it without trouble, and all of my devices were detected and drivers provided during the install.
I wanted to upload the Windows Server 2008 R2 ISO onto it so I could build a VM. The ISO is only 3GB so I figured it should only take a few minutes to copy. Instead its taking close to 30 minutes, and I'm seeing about 5000 KBps of write throughput in the performance monitor. The volume in question is a RAID1 mirror of two 15Krpm SAS 72GB HDDs, so I should be getting something like the maximum typical performance for that drive, which should be significantly higher.
What's going on here?
Does the controller have the cache battery module installed? This server is old enough that the original battery may dead and or removed. If it is installed and working is write caching enabled? Failure to have write caching enabled will seriously affect performance.
Well the BBU is there (I physically inspected it), and it was happy until I took the box out of service a couple of months ago after having moved its Windows workload onto another VMWare box (and its been powered off since). The current version of the firmware doesn't give me any options in the BIOS except to create/view/delete an array, and tells me to use the Windows-based utilities for more options. So I can't even see how its set. Although when I set the box up initially I recall the disk performance being very fast.
I'm downloading the Proliant firmware DVD now, plus the update for the P600. I'll see how things look after I upgrade all of the various bits of microcode.
Perhaps being powered off for so long has exhausted the BBU? It had line power, but wasn't turned on. Would a drained BBU cause the write cache to not work until the BBU was charged? There's no way of knowing since I don't have the windows tools running on the hardware to tell me about it.
Download the Smartstart CD too. This contains the ACU (Array Configuration Utility) which displays the status of the battery. Unless the battery is fully charged the write cache is disabled. In this case you should also see a message in during boot up for the RAID controller.
Well it definitely isn't complaining about any array error or warning conditions at boot, and I've seen those before (e.g. "valid data found in array accelerator cache, BBU is dead, etc."). I'm still waiting for the USB key formatter to finish doing its work (it seems to run pretty slowly for me). I do have SmartStart, so I checked that out, and the write cache was disabled, so I enabled it.
I noticed before I shut the box down that the VMWare system boot volume (the first pair of 10K SAS drives) had both drive lights blinking alternately like it was doing some sort of sync (I didn't run the ADU alas), and it stopped after a few minutes. Thinking that maybe this was the cause of the performance issue, I tried copying the same 600MB file to both datastore2 and datastore1 separately. The copy to DS2 went at about 15MBps, while the copy to DS1 went at the same 5MBps. Doing them at the same time yielded the result that the total max throughput to both volumes was 15MBps.
I should be seeing around 80MBps for a typical 10K SAS drive (I mistakenly thought they were 15K) without the storage controller. All I'm doing is mirroring.
Now this is just doing the upload through the datastore browser - I haven't created any VMs yet or done any real testing, so is this just an artifact of the datastore browser upload? Or do I have some wierd issue here?