Nimble HF20 Upgrade Option

[Dylan-G]

We have a Nimble HF20 with 6x960GB SSD. Current we have
CPU saturation over 200% and
Cache saturation only 36%
Latency Max is 16ms
Thoughput max is 345MiBps

For we have no significant impact on performance yet, but we do want a time bomb with us.

So what are our upgrade option?

Is controller upgrade from HF20 to HF40 is zero downtime one by one hotswap?

Cache upgrade looks like one by one swap to large size SSD, right?

Can we reuse the old cache SSD in capacity tier? We still have slot there.

Hope you can give your experience.

Thanks/
Dylan

[Nick_Dyer]

Hello, thanks for posting the question.

Nimble arrays are able to live-swap components (controllers and SSDs) in order to upgrade for various reasons.

If your CPU staturation is >200% according to Infosight, then a set of HF40 or HF60 controllers would be your applicable upgrades. This will increase your IOPS and throughput capability. 

If Infosight is only reporting cache utilisation at 36%, then you don't really need to swap them out for bigger SSDs. You would only do this typically if you wanted to pin workloads in the cache, or if you wanted to expand the working set of data on the array. Note: adding a capacity expansion shelf also adds SSD cache to the array automatically.

Infosight will recommend what upgrade is the right move, under the "executive dashboard" page

[Dylan-G]

Thanks Nick.

How do we decide whether upgrade to HF40 or HF60?

Since we still have plenty of capacity left, does turn off the compression and dedup help to reduce the cpu utilization?

Does it means we can pin about 4TB volume in cache?

Thanks/
Dylan

[Nick_Dyer · Accepted solution]

Hello,

Take a look under "Executive Dashboard" on Infosight. At the bottom you'll find "upgrade recommendations" - where we direct you to exactly what you should be thinking to upgrade - array CPU, cache or capacity.

There's no benefit for disabling compression on Nimble - it has no CPU overhead. You could disable dedupe if you wanted to, and you may see 5-10% CPU utilisation benefit.

Finally, you can only ever pin 45% of the overall usable cache capacity - in order to ensure you don't overrun the cache and detriment non-pinned workloads (remember, it's a cache, not a tier of storage). I wrote a blog post on cache pinning many years ago: https://community.hpe.com/t5/HPE-Storage-Tech-Insiders/Nimble-OS-2-3-Implementing-All-Flash-Service-Levels/ba-p/6986413. It's for certain workloads which you want to guarantee all-flash performance. Note: pinning workloads is not possible on volumes that are deduplicated.

[Dylan-G]

Thanks Nick.

Great "insight" !