Read latency

[Fizzer]

Hi All, my first post here so please excuse any innacuracies! I've had our EF2824 for a few days and have configured it, a 10GB switch and a new R740 esxi 6.5 host. It's hosting just a single 2TB datastore. It's all SSD.

I've spun up a windows server 2012 VM for performance testing on the host. Connectivity seems fine and at the moment it's just doing updates - nothing too heavy. onceit's finished I was intending to install benchmarking tools to check performance.

However I have noticed that , even while it's just prepping and write latency is fantastic, read latency is averaging about 50ms so I'm guessing that something is wrong...

I have configured jumbo frames (9000) throughout.

Attached are the performance stats from the device, controller B is taking most of the load and there are 2 latency values for each controller channel:

Controller B Drive Channel 1 (Internal) - seems fine (<2ms)

Controller B Host Channel 1 - abysmal (>60ms)

Does this indicate a network problem if the internal figures are good and (presumably) external figures are bad?

If so, how come write rates are normal? surely a network issue would be bi-directional?

I know I'll eventually get this sorted but any pointers would be gratefully received!

Kind regards

Andy

[DannyL · Accepted solution]

Hi Andy,

I am going to refer you to the following article: https://kb.netapp.com/app/answers/answer_view/a_id/1074155/loc/en_US . With low queue depth / heavy read workloads, we recommend disabling the Delayed ACK setting to resolve the issue you may be facing.

Thanks,

Danny

[Fizzer]

Hi Danny,

I meant to reply to this earlier today, that was indeed the problem. I disabled delayed ack on my host and it's sorted the issue.

Atto disk benchmark in a simple windows server is giving me some great throuput now.

Thanks for your help

Andy