VMware Cloud Community
jvm2016
Hot Shot
Hot Shot

cache_consideration_in disk group_vsan

Hi Bob/Vsan team ,

can you please suggest the best practice for cache in configuring disk group for vsan.

below is what iam trying to configure .

4 esxihost vsan cluster (uniform resources) with one disk group each .

for disk group i have 4 capcity hard disk  of 8TB.

and ssd for cache 800GB.

is this sufficient cache ?? my understading that cache/capacity shud be at least 10 percent per disk group.

Please suggest.

Reply
0 Kudos
6 Replies
TheBobkin
Champion
Champion

Hello,

Do you mean 4x8TB HDDs or 4x2TB HDDs (8TB total)? The former would be 2.5% cache:capacity ratio which is way below guidelines and would likely have low performance, the latter (10% ratio) is okay.

Bob

Reply
0 Kudos
jvm2016
Hot Shot
Hot Shot

4 hdd of 8 TB each .and 800 GB ssd as cache.

Reply
0 Kudos
TheBobkin
Champion
Champion

As I said above - that is a fraction of the advised cache:capacity ratio and I wouldn't advise configuring it like this unless you are 100% positive the intended workload is something that has very low storage throughput/performance requirements and is not latency-sensitive (e.g. some form of predominantly static archival data).

VMware recommendations for cache:capacity ratio isn't just an arbitrary number but based on averages ratio % of warm data - if you have a workload that is significantly lower than ~10% warm at any one time then potentially less than 10% cache won't have a noticeable impactful (e.g. high latency at the VM/Application level). But if you have 10% or more data warm this won't perform well as your cache devices will be saturated and the slower large capacity devices will be unable to accept destaged writes fast enough to keep up.

Also there can be relatively large differences in performance capabilities between lowest and higher end HDDs e.g. 7.2k SATA vs 15k SAS so just going by size isn't ideal - in general you get what you pay for so buying cheaper disks isn't always a good course of action (but as I said, it depends on the intended use and expectations/requirements).

Bob

Reply
0 Kudos
jvm2016
Hot Shot
Hot Shot

thnaks.I am checking this  agian .

iam feeding rvtoolsreport to vsan calculator without being worried about of warm data.

Reply
0 Kudos
TheBobkin
Champion
Champion

Just to clarify further (as it seems with the passage of time I am slowly forgetting Hybrid stuff :smileygrin: 😞 the advisement of '10%' is based on usable space (e.g. without replicas/mirrors), so a FTM=RAID1, FTT=0/1/2 vmdk with 100GB used would all be 10GB advised cache. This gets slightly more complicated when you start to consider a thin vmdk provisioned 100GB isn't necessarily using all its available space but if it grows over time it will use more and thus cache:capacity ratio will decrease (as available cache remains the same but the capacity usage increases).

"iam feeding rvtoolsreport to vsan calculator without being worried about of warm data."

While I am aware it takes time and isn't always possible, one should always aim to get more specifics insight into what fits the bill here with regard to the intended usage (e.g. is the workload pure SQL, VDI, 'mixed'), this can be achieved by directly categorising and analysing the current workload(s) and preferably by working with the current owners/administrators/users who *should* have some insights.

Bob

Reply
0 Kudos
jvm2016
Hot Shot
Hot Shot

ThnaksI ma checking this.

Reply
0 Kudos