Hi List, I am working towards getting some performance number for openstack Cinder when backed with GlusterFS as the storage backend. I am using pblio [1] workload generator tool for the same.
In short, Pblio is a synthetic OLTP <http://en.wikipedia.org/wiki/Online_transaction_processing> enterprise workload used to stress storage systems. This benchmark will stress a storage system to determine the maximum number of IOPS it can manage before having a mean response latency of 30 milliseconds or greater. More details are available @ [1] Currently I just ran pblio in a *mock setup*, details of the setup are below. More real-world setup numbers will come soon. *Setup:* I have my host laptop, devstack AIO runs in a VM and inside that is my nova VM (nested VM thus) where I am running pblio. Nested VM runs F22. FWIW, my laptop has only SSDs so everything is SSD backed. *Screenshots:* [stack@devstack-f21 ~]$ [admin] ssh -i ./mykey.pem [email protected] "lsblk" NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINT sr0 11:0 1 418K 0 rom vda 252:0 0 20G 0 disk └─vda1 252:1 0 20G 0 part / *vdb 252:16 0 3G 0 disk vdc 252:32 0 3G 0 disk vdd 252:48 0 1G 0 disk * * So vdb, vdc and vdd are the glusterfs backed block devices in my nova VM Some of the pblio runs that I did ... [root@vm1 pblio]# ./pblio -asu1=/dev/vdb -asu2=/dev/vdc -asu3=/dev/vdd -runlen=60 -bsu=2 ----- pblio ----- Cache : None ASU1 : 3.00 GB ASU2 : 3.00 GB ASU3 : 0.67 GB BSUs : 2 Contexts: 1 Run time: 60 s ----- Avg IOPS:100.57 Avg Latency:0.8021 ms [root@vm1 pblio]# ./pblio -asu1=/dev/vdb -asu2=/dev/vdc -asu3=/dev/vdd -runlen=600 -bsu=60 ----- pblio ----- Cache : None ASU1 : 3.00 GB ASU2 : 3.00 GB ASU3 : 0.67 GB BSUs : 60 Contexts: 1 Run time: 600 s ----- Avg IOPS:3000.48 Avg Latency:6.1536 ms [root@vm1 pblio]# ./pblio -asu1=/dev/vdb -asu2=/dev/vdc -asu3=/dev/vdd -runlen=600 -bsu=100 ----- pblio ----- Cache : None ASU1 : 3.00 GB ASU2 : 3.00 GB ASU3 : 0.67 GB BSUs : 100 Contexts: 1 Run time: 600 s ----- Avg IOPS:4831.87 Avg Latency:33.9057 ms * So somewhere close (on the lesser side) to (100 * 50) is the max IOPS for this setup, given the above data This is very very draft/mock run, no where close to real world setup I wanted to share this with the broader audience, comments welcome. More updates will come, once I run these tests on a larger close-to-real-world setup. thanx, deepak
_______________________________________________ Gluster-devel mailing list [email protected] http://www.gluster.org/mailman/listinfo/gluster-devel
