• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

SSD MS Latency vs IOPS

IOPS is just transfer rate.

You want to know at what transfer rate latency goes up? that's complicated depending on what you are doing (read/write - erase/trim/etc).

What are you actually trying to work out?
 
We have a server side application that has fairly tight tolerances to read latency, we need reads to be always under 5ms, 10ms at the absolute worst. And given x number of servers with X number of drives how many IOPS can we handle before latency falls out of tolerances.

I'm working on setting up a test bed now, but if I had a graph that showed read latency vs IOPS just for single SSD it would help a lot in knowing what we could handle and help me frame my test. IE:

if a normal SSD's latency scales linearly as IOPS increase it would be different from having it be an exponential graph where they can handle a lot of IOPS at a specific latency and then there being a threshold where latency increases very quickly.
 
What you are saying makes no sense.

Latency is a measure of time. (so it can be expressed in units like s, ms, etc.)
IOPS is a frequency . (so it can be expressed in units of Hz, kHz, MHz, etc).

I think maybe you are confusing IOPS with queue depth (which for the synthetic tests, queue depth should be the same as # of test threads).

So if anything, latency and IOPS are going to be INVERSELY proportional.
Assuming the test runs as fast as possible on each thread (why wouldn't it?),
latency = QD (queue depth or thread count) / IOPS (assuming you are throwing as many IOPS at it as possible limited by the response time and your thread count).
 
Back
Top