I ran through the iometer tests on the vmktree iso and below are the results from an NFS share. The first result is with the NFS share cached which is the wrong behavior. The second test being after the MR1 install with the cache turned off. On the Max Throughput-50% read it is taxing the 1Gb connection now. My numbers are not overly scientific given that there are 20 VM's running on the 3100 as I ran the test but the performance difference is substantial for those of you fighting with NFS performance.
I was planning on rolling out the VNXe and my new hosts this month, but until I can get some decent numbers out of this thing or can figure out what I am doing wrong, it's just and expensive space heater. I am working with EMC tech support, but so far, I have not seen any performance improvement. Even simple file copies within the VM seem slow. For example, went to a web share on my network that has a bunch of installation files on it. I tried to copy a file that was about 8 MB from the network share to the desktop of my VM and the download transfer rate was maxed out at 200KBps before settling at 150 KBps. It took 46 seconds to copy this one file from a physical machine to the virtual desktop.
I am no storage expert, which is why I purchased the VNWe by the way, so I'm sure that I am doing something wrong, but have not figured out what. If I find out anything from Tech Support, I will post it here.
Did you run your own IOmeter stats? If so, what is your setup and what were your results?
My VNXe implementation runs without "glory". I've implemented it for 15 VMs (40 Users). All services are virtualized (exchange 2007, DCs, MSCRM, ....) during the daily workloads all work without problems, but when I use massive robocopy, sVmotion or somethings else that are disk intensive the VNXe performance falls down. The read or write latency (see them in vCEnter) are realy bad, about 600-700 ms of latency.
I've just opened a SR but they told me that the configs are all ok and there are no problem. I'm quite unsatisfied of the box. I think that this type of storage MUST be configured with many disks....the best way is to buy it with 12 SAS 15 K disks, to fill the first dae.
scamp2
2 Posts
3484
1
Posted July 20th, 2011 15:00
I ran through the iometer tests on the vmktree iso and below are the results from an NFS share. The first result is with the NFS share cached which is the wrong behavior. The second test being after the MR1 install with the cache turned off. On the Max Throughput-50% read it is taxing the 1Gb connection now. My numbers are not overly scientific given that there are 20 VM's running on the 3100 as I ran the test but the performance difference is substantial for those of you fighting with NFS performance.