Messages in this thread | | | Subject | Re: [PATCH 1/9] move blk_iopoll to limit and make it generally available | From | Sagi Grimberg <> | Date | Sun, 15 Nov 2015 10:48:41 +0200 |
| |
>> On Fri, Nov 13, 2015 at 3:46 PM, Christoph Hellwig <hch@lst.de> wrote: >>> The new name is irq_poll as iopoll is already taken. Better suggestions >>> welcome. >> >> Sagi (or Christoph if you can address that), >> >> @ some pointer over the last 18 months there was a port done at >> mellanox for iser to use blk-iopoll and AFAIR it didn't work well or >> didn't work at all. Can you tell now what was the problem and how did >> you address it at your generalization? > > Hi Or, > > Sagi mentioned last time he tried a similar approach in iSER he saw > some large latency sparks. We've seen nothing worse than the original > approach. The Flash memory summit slide set has some numbers: > > http://www.flashmemorysummit.com/English/Collaterals/Proceedings/2015/20150811_FA11_Bandic.pdf > > they aren't quite up to date, but the latency distribution hasn't > really changed.
Or is correct,
I have attempted to convert iser to use blk_iopoll in the past, however I've seen inconsistent performance and latency skews (comparing to tasklets iser is using today). This was manifested in IOPs test cases where I ran multiple threads with higher queue-depth and not in sanitized pure latency (QD=1) test cases. Unfortunately I didn't have the time to pick it up since.
I do have every intention of testing it again with this. If it still exist we will need to find the root-cause of it before converting drivers to use it.
Sagi.
| |