Dear all,
I am having serious issues on transcoding speed and after spending considerable time in troubleshooting the issue, I am starting to run out of ideas so your help would be highly appreciated.
I was planning on moving our Plex installation from a physical server (i5-3570K, 16GB, Windows 8.1) to a virtual machine in ESXi 6.0 host. The host has two Xeon E5-2650's (total 16 cores / 32 HT) and plenty of memory. For comparison, the CPU passmark scores are i5 7150 vs Xeon 2 x 10394 (although single thread performance is lower).
Creating Ubuntu 14.04 server with 8 vCPUs and 32GB memory I was hoping to achieve similar performance than on the previous physical server. Now the reality is that when I am transcoding content to be synched to iOS device (easiest way to benchmark), the new installation is giving maximum transcoding speed of x4.5 to 5 compared to the previous physical server doing the same with x35-40 speed. Quite a difference!
To troubleshoot the issue I have tried at least the following:
* Update PMS from 0.9.14.4 to 0.9.15.0 (Plex pass) - no effect
* Setup Windows 2012 R2 (std) Server with similar virtual hw - exactly similar results.
* Increase amount of memory from 4 to 32GB - no effect
* No obvious errors in logs
* As media is on NAS, copy media to local drive and create new library for it - no effect
* Move Plex Cache to ram disk (tmpfs) - no effect
* Move whole Plex library (/var/lib/...) to ram disk - no effect
* Change transcoding settings to "Make my CPU hurt" - increases CPU utilisation but not speed
As the virtual CPUs are not fully loaded (htop or vSphere client measurements) after the vCPU count is 4 or higher, I am fairly sure that something else is limiting the performance but I am fast running out of ideas to try.
I would really appreciate good suggestions to try, either in the VM or on the ESXi side. Thanks already in advance!
Regards,
Tom
