I have a ESXi 5.5 server running with a FC SAN. The FC san is created using an openfiler OS with software RAID and exported with the qlatgt driver. Direct writes to this SAN when no VMFS is used work fine at full speed (>100MB/s).
Virtual machines run on a VMFS 5 datastrore that has been allocated on the FC SAN.
However, the virtual machines that are on the datastore have extremely slow disk speeds (10MB/s). The VMs are Centos 6.5 64bit 2.6.32 kernel.
To identify the bottleneck, I created a separate LUN on my filer to find the problem with only 1 disk (non RAID, direct block export of 1 disk).
When I create a datastore (VMFS5) on the new LUN, create a virtual disk on it, and mount it in a linux VM, dd tests show 10MB/s write speed of /dev/zero to this virtual disk on the new LUN's datastore. When I pass the LUN to the VM using RDM (raw device mapping, so no VMFS or virtual disks inbetween), the speed goes to 110MB/s (physical disk speed).
Disk writes up to 5G are tested, all similar.
This test locates the problem in the ESXi and not down in my FC SAN (eg. switch, filer, disks...). Something goes wrong in the datastore/virtual disks. I tried changing in de virtual disks between thick and thin provisioning but nothing changes. I actually want to use the ESXi datastore.
Why is the performance via the virtual disks and VMFS5 datastore so slow. Whan can I change in my configuration to chang it?
Thank you
Jeffrey