On 12/29/19 4:16 PM, Sitsofe Wheeler wrote:
(CC'ing Vincent for comment)
On Thu, 19 Dec 2019 at 17:34, <Charles.Colburn@xxxxxxxx> wrote:
Sitsofe
Is there an example of using the steadystate option to do random preconditioning of and SSD? - I assume that is what it is for. I tried to use the featured a while back but never got it to work as expected.
Chuck Colburn
-----Original Message-----
From: fio-owner@xxxxxxxxxxxxxxx <fio-owner@xxxxxxxxxxxxxxx> On Behalf Of Sitsofe Wheeler
Sent: Thursday, December 19, 2019 1:11 AM
To: Mike Kingsbury
Cc: fio
Subject: Re: Running FIO until DUT leaves steadystate
[EXTERNAL EMAIL]
Hi,
On Fri, 13 Dec 2019 at 16:29, Mike Kingsbury <mike.kingsbury@xxxxxxxxxxx> wrote:
Reading through the docs and looked at some examples, its not clear if
its possible to accomplish this. In our case, we're attempting to
evaluate SSDs and determine when they start internal garbage
collection and how much performance drops. Is there a way to detect
steadystate, and then exit when the DUT performance leaves that
steadystate?
Not really. You can use the steadystate option (https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-steadystate
) to try and determine when you're in the steadystate but then you're kind of stuck. Maybe once you knew what expected latencies were, you could then use max_latency (https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-max-latency
) to allow a job to exit if one got obviously too high? You would have to do a fair bit of intermediate storage and processing to make this work and it's not clear how reliable it would be though...
Chuck, I don't understand why one would need to engage steady state
detection to randomly precondition an SSD. Why not just use an LFSR to
write the device's logical capacity? Is there something you mean by
"random preconditioning" that isn't being communicated to me?
Mike, as for your original question, I do not believe that the
steadystate feature is suitable for detecting when garbage collection
has started. Garbage collection presumably will begin at the point where
the total amount of write IO approaches the device's physical capacity.
Vincent