[Gluster-users] High CPU utilization on Windows Guest OS with random write workload when using Cinder Volume (libgfapi)

[qingwei wei]

Hi all,

I am currently testing running Openstack instance on Cinder volume with
libgfapi. This instance is the Windows instance and i found that when
running random 4k write workload, the CPU utilization is very high,  90%
CPU utilization with about 86% in privileged time. I also tested the
workload with volume from NFS and the CPU utilization is only around 5%. Is
there any reason why is that so? Appreciate on any comments and suggestions.

Details.

Workload generator: IOMeter
Instance: Windows 2008R2
Openstack Host: CentOS 7
Workload: 4kb random write, 4 workers with each worker configure to have 4
outstanding IOs
Gluster server and client version: 3.7.8

Gluster info
Volume Name: g37test
Type: Distribute
Volume ID: 27947421-111b-4727-bcaf-bafe73debe07
Status: Started
Number of Bricks: 1
Transport-type: tcp
Bricks:
Brick1: 192.168.99.4:/mnt/ssd/g37test
Options Reconfigured:
performance.readdir-ahead: on
nfs.disable: true
performance.io-cache: off
performance.quick-read: off
server.allow-insecure: on
storage.owner-uid: 165
storage.owner-gid: 165


IOmeter result:

Glusterfs
======
IOPS: ~5800
CPU utilization: ~90%

NFS
===
IOPS: ~6000
CPU Utilization: ~5-6%



Cw
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20160308/c7140f78/attachment.html>