Hi Paul, not helping directly answering your question, but here are some observations and "warning" if client's are using write-back cache on KVM level
I have (long time ago) tested performance in 3 combinations (this was not really thorough testing but a brief testing with FIO and random IO WRITE) - just CEPH rbd cache (on KVM side) i.e. [client] rbd cache = true rbd cache writethrough until flush = true #(this is default 32MB per volume, afaik - just KMV write-back cache (had to manually edit disk_offering table to activate cache mode, since when creating new disk offering via GUI, the disk_offering tables was NOT populated with "write-back" sertting/value ! ) - both CEPH and KVM write-back cahce active My observations were like following, but would be good to actually confirm by someone else: - same performance with only CEPH caching or with only KVM caching - a bit worse performance with both CEPH and KVM caching active (nonsense combination, I know...) Please keep in mind, that some ACS functionality, KVM live-migrations on shared storage (NFS/CEPH) are NOT supported when you use KVM write-back cache, since this is considered "unsafe" migration, more info here: https://doc.opensuse.org/documentation/leap/virtualization/html/book.virt/cha.cachemodes.html#sec.cache.mode.live.migration or in short: " The libvirt management layer includes checks for migration compatibility based on several factors. If the guest storage is hosted on a clustered file system, is read-only or is marked shareable, then the cache mode is ignored when determining if migration can be allowed. Otherwise libvirt will not allow migration unless the cache mode is set to none. However, this restriction can be overridden with the “unsafe” option to the migration APIs, which is also supported by virsh, as for example in virsh migrate --live --unsafe " Cheers Andrija On 20 February 2018 at 11:24, Paul Angus <paul.an...@shapeblue.com> wrote: > Hi Wido, > > This is for KVM (with Ceph backend as it happens), the API documentation > is out of sync with UI capabilities, so I'm trying to figure out if we > *should* be able to set cacheMode for root disks. It seems to make quite a > difference to performance. > > > > paul.an...@shapeblue.com > www.shapeblue.com > 53 Chandos Place, Covent Garden, London WC2N 4HSUK > @shapeblue > > > > > -----Original Message----- > From: Wido den Hollander [mailto:w...@widodh.nl] > Sent: 20 February 2018 09:03 > To: dev@cloudstack.apache.org > Subject: Re: Caching modes > > > > On 02/20/2018 09:46 AM, Paul Angus wrote: > > Hey guys, > > > > Can anyone shed any light on write caching in CloudStack? cacheMode is > available through the UI for data disks (but not root disks), but not > documented as an API option for data or root disks (although is documented > as a response for data disks). > > > > What hypervisor? > > In case of KVM it's passed down to XML which then passes it to Qemu/KVM > which then handles the caching. > > The implementation varies per hypervisor, so that should be the question. > > Wido > > > > #huh? > > > > thanks > > > > > > > > paul.an...@shapeblue.com > > www.shapeblue.com > > 53 Chandos Place, Covent Garden, London WC2N 4HSUK @shapeblue > > > > > > > -- Andrija Panić