blkif sort queue

blkif sort queue

[Eric Mowat]

Hi, I'm one of the engineers working on the FreeBSD port and while 
upgrading to TOT, I noticed that there appears to only be 1 sort queue 
for all guest domain devices (xlbd_blk_queue).  All disks end up 
pointing at this request queue.  The strategy call (do_blkif_request) 
uses this global queue and attempts to dequeue using elevator sort.  
Does any kind of sort make sense given that multiple devices feed this 
queue?

I ask because some of the new recovery code attempts to call 
kick_pending_request_queues() in blkif_connect().  In my FreeBSD version 
prior to this I was using per disk sort queues, which was fine since I 
would only try to kick the i/o on interrupt completion.  In this newer 
context it seems like having only 1 queue avoids the need to keep a 
table of per-disk sort queues. 

Thanks for any help.

    Eric


-------------------------------------------------------
This SF.net email is sponsored by: IT Product Guide on ITManagersJournal
Use IT products in your business? Tell us what you think of them. Give us
Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more
http://productguide.itmanagersjournal.com/guidepromo.tmpl

Re: blkif sort queue

[Keir Fraser]

The scheduling in the frontend domain probably doesn't buy us very
much as the backend will do its own scheduling, and has a better idea
about the mapping of block addresses onto spindles. 
However, the extra overhead is not going to hurt our disc
performance. 

 -- Keir


> Hi, I'm one of the engineers working on the FreeBSD port and while 
> upgrading to TOT, I noticed that there appears to only be 1 sort queue 
> for all guest domain devices (xlbd_blk_queue).  All disks end up 
> pointing at this request queue.  The strategy call (do_blkif_request) 
> uses this global queue and attempts to dequeue using elevator sort.  
> Does any kind of sort make sense given that multiple devices feed this 
> queue?
> 
> I ask because some of the new recovery code attempts to call 
> kick_pending_request_queues() in blkif_connect().  In my FreeBSD version 
> prior to this I was using per disk sort queues, which was fine since I 
> would only try to kick the i/o on interrupt completion.  In this newer 
> context it seems like having only 1 queue avoids the need to keep a 
> table of per-disk sort queues. 
> 
> Thanks for any help.
> 
>     Eric
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: IT Product Guide on ITManagersJournal
> Use IT products in your business? Tell us what you think of them. Give us
> Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more
> http://productguide.itmanagersjournal.com/guidepromo.tmpl
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.net email is sponsored by: IT Product Guide on ITManagersJournal
Use IT products in your business? Tell us what you think of them. Give us
Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more
http://productguide.itmanagersjournal.com/guidepromo.tmpl