Open Stack

Hi,

I think that https://wiki.mikejung.biz/KVM_/_Xen might be quite useful
Pretty much there won’t be a universal recipe but rather combination of what are kernel/QEMU/OpenStack… versions
and implementation choices that were made
-Guest OS/host OS kernel version
-QEMU version
-OpenStack release. Basically the part of XML definition of disk is controlled by OpenStack
-virtio_blk vs. some fully-virtualized stuff like ide is used
-the backend for virtual disk

some points possibly to add to that article:

1)    For virtio_scsi in OpenStack see https://blueprints.launchpad.net/nova/+spec/cinder-backend-report-discard



2)    Linux itself is supposed to detect if it works on HDD or SSD



See:

smartctl -a /dev/sd<x>

and

cat /sys/block/sd<x>/queue/rotational

(0 –SSD, 1 – HDD)

But results might be contradictory/interpretation of that is not straightforward.

Maybe someone has better insight, but what I understood from one hand not all SSD drives report to being as such

And that’s why sysfs entry /sys/block/sd<x>/queue/rotational was introduced which was meant to enforce ‘SSD-like behavior)

https://lwn.net/Articles/408428/

But then I kinda understood that Linux might/is supposed to set it correctly (via udev rules?)

But it should be noted in Ubuntu 12.04/Ubuntu 14.04 that I worked with that had SSDs this flag /rotational was ‘1’ on SSD, ie. Incorrect…)

3)     The question of cfq vs. deadline vs. noop scheduler (apparently both in guest and host) where decision should be based on workloads/recommendations of OS vendor (/which again might be release-dependent).


BR.
Konstantin



From: Tim Bell [mailto:Tim.Bell at cern.ch]
Sent: Friday, August 05, 2016 12:44 PM
To: openstack-operators <openstack-operators at lists.openstack.org>
Subject: [Openstack-operators] Tuning I/O with SSDs


Has anyone a good recipe for improving I/O performance when the hypervisor has SSDs ?

The configuration is CentOS 7 for guest and hypervisor with KVM.

Tim

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20160805/082b5391/attachment.html>