NVIDIA
Capture vGPU performance, Host vCPU, VM vCPU and memory performance using rrd2csv
Intent: Capture the virtual machines memory, vCPU performance data along with the vGPU's GPU utilization in the hypervisor. Get a list of the virtual machines using: [root@nvgrid ~]# xe vm-list | awk '{if ( $0 ~ /uuid/) {uuid=$5} if ($0 ~ /name-label/) {$1=$2=$3="";vmname=$0; printf "%20s - %s\n", vmname, uuid}}' hp2-vgpu1 - 8c8781a1-c8eb-3471-2f19-37c68407015e hp2-vgpu2 - 02ba1ffc-c8b1-8fca-a54b-edae4902a5d1 hp2-vgpu3 - 3061703c-374a-4fc5-8b65-23a84cd01106 hp2-vgpu4 - 9034ca98-fee3-643f-fc95-f189e42cb4c4 Set environment variables for each of the virtual machines to make the rrd2csv a 'little' more readable: [root@nvgrid ~]# export UUID1=8c8781a1-c8eb-3471-2f19-37c68407015e [root@nvgrid ~]# export UUID2=02ba1ffc-c8b1-8fca-a54b-edae4902a5d1 [root@nvgrid ~]# export UUID3=3061703c-374a-4fc5-8b65-23a84cd01106 [root@nvgrid ~]# export UUID4=9034ca98-fee3-643f-fc95-f189e42cb4c4 [root@nvgrid ~]# rrd2csv -n -s 5 AVERAGE:vm:$UUID1:memory AVERAGE:vm:$UUID1:cpu AVERAGE:vm:$UUID2:memory AVERAGE:vm:$UUID2:cpu AVERAGE:vm:$UUID3:memory AVERAGE:vm:$UUID3:cpu AVERAGE:vm:$UUID4:memory AVERAGE:vm:$UUID4:cpu AVERAGE:host::cpu_avg AVERAGE:host::gpu > ./vGPU-VMs-perf.csv Now connect to your virtual machines and run some benchmarks and/or applications. This rrd2csv query utilizes wildcards by not being specific about all counter source names. If you wish to capture specific counters use the direct name following the AVERAGE:host::cpu_avg example. <CTRL + c> to stop the capture process, then transfer the resulting .csv file to view the results. Some counters will not have data, please look closely at the rrd2csv help via rrd2csv -help as well as : Performance Monitoring Enhancements Pack for XenServer 6.1 [url]http://support.citrix.com/article/CTX135033[/url]
Intent:
Capture the virtual machines memory, vCPU performance data along with the vGPU's GPU utilization in the hypervisor.

Get a list of the virtual machines using:
[root@nvgrid ~]# xe vm-list | awk '{if ( $0 ~ /uuid/) {uuid=$5} if ($0 ~ /name-label/) {$1=$2=$3="";vmname=$0; printf "%20s - %s\n", vmname, uuid}}'
hp2-vgpu1 - 8c8781a1-c8eb-3471-2f19-37c68407015e
hp2-vgpu2 - 02ba1ffc-c8b1-8fca-a54b-edae4902a5d1
hp2-vgpu3 - 3061703c-374a-4fc5-8b65-23a84cd01106
hp2-vgpu4 - 9034ca98-fee3-643f-fc95-f189e42cb4c4

Set environment variables for each of the virtual machines to make the rrd2csv a 'little' more readable:
[root@nvgrid ~]# export UUID1=8c8781a1-c8eb-3471-2f19-37c68407015e
[root@nvgrid ~]# export UUID2=02ba1ffc-c8b1-8fca-a54b-edae4902a5d1
[root@nvgrid ~]# export UUID3=3061703c-374a-4fc5-8b65-23a84cd01106
[root@nvgrid ~]# export UUID4=9034ca98-fee3-643f-fc95-f189e42cb4c4

[root@nvgrid ~]# rrd2csv -n -s 5 AVERAGE:vm:$UUID1:memory AVERAGE:vm:$UUID1:cpu AVERAGE:vm:$UUID2:memory AVERAGE:vm:$UUID2:cpu AVERAGE:vm:$UUID3:memory AVERAGE:vm:$UUID3:cpu AVERAGE:vm:$UUID4:memory AVERAGE:vm:$UUID4:cpu AVERAGE:host::cpu_avg AVERAGE:host::gpu > ./vGPU-VMs-perf.csv

Now connect to your virtual machines and run some benchmarks and/or applications. This rrd2csv query utilizes wildcards by not being specific about all counter source names. If you wish to capture specific counters use the direct name following the AVERAGE:host::cpu_avg example.

<CTRL + c> to stop the capture process, then transfer the resulting .csv file to view the results.

Some counters will not have data, please look closely at the rrd2csv help via
rrd2csv -help

as well as :
Performance Monitoring Enhancements Pack for XenServer 6.1
http://support.citrix.com/article/CTX135033

Jeremy Main
Lead Solution Architect - NVIDIA GRID
GRID Resources : http://www.nvidia.com/object/grid-enterprise-resources.html
GPUProfiler : http://gpuprofiler.com/

#1
Posted 05/19/2014 02:15 PM   
Thanks Jeremy, my white paper on metrics has been wrapped into main doc, Sally our user experience lead did a great job adding a new chapter 9 with loads of other useful metrics, vGPU should join this in the next main release http://blogs.citrix.com/2014/01/22/xenserverxendesktop-vgpu-new-metrics-available-to-monitor-nvidia-grid-gpus/ A lot of metrics of by default but some super storage and IO Nvidia now also have the NVWMI toolkit link on this page http://blogs.citrix.com/2014/03/31/citrix-and-nvidia-extend-vgpu-support-matrix-to-include-windows-8-1-and-windows-server-2012-r2/
Thanks Jeremy, my white paper on metrics has been wrapped into main doc, Sally our user experience lead did a great job adding a new chapter 9 with loads of other useful metrics, vGPU should join this in the next main release http://blogs.citrix.com/2014/01/22/xenserverxendesktop-vgpu-new-metrics-available-to-monitor-nvidia-grid-gpus/

A lot of metrics of by default but some super storage and IO

Nvidia now also have the NVWMI toolkit link on this page http://blogs.citrix.com/2014/03/31/citrix-and-nvidia-extend-vgpu-support-matrix-to-include-windows-8-1-and-windows-server-2012-r2/

#2
Posted 05/28/2014 08:30 PM   
Scroll To Top

Add Reply