AFF, NVMe, EF-Series, and SolidFire Discussions

Total IOPS more than expected




configuration: AFF320, FC, 4 node cluster with  NVMe SSD, multiple SVMs(each SVM is a different customer)


Recently while testing for noisy neighbors, our platform guys realized that they were getting 400K IOPS from LUNs assigned from 4 volumes on 2 nodes in the cluster. Expectation from 2 nodes was a max of 250K IOPS with 100% read and 8K blocks, without any QOS applied.


Our understanding was 4 Node NetApp cluster with 12 SSDs/HA pair would yield max 440K IOPS. Is that wrong? From the tests, it seems with only 2 nodes no QOS policy, we can go beyond 400K and that would impact other volumes on the cluster.


Is my understanding correct?


Re: Total IOPS more than expected


It is all workload dependent. If the workload is mostly reads, and specifically, re-reads that are in cache, you can certainly go way beyond the limit.


You do have the option of imposing QoS limits, both floor and ceiling on the AFF platform which artificially limits to the clients

Earn Rewards for Your Review!
GPI Review Banner
All Community Forums