On Wed, Mar 26 2003, Badari Pulavarty wrote:
> On Tuesday 25 March 2003 04:35 am, Jens Axboe wrote:
>
> > Only testing will tell, so yes you are very welcome to give it a shot.
> > Let me release a known working version first :)
>
> Jens,
>
> I found whats using 32MB out of 8192-byte slab.
>
> size-8192 before:10 after:4012 diff:4002 size:8192 incr:32784384
>
> It is deadline_init():
>
> dd->hash = kmalloc(sizeof(struct list_head)*DL_HASH_ENTRIES,GFP_KERNEL);
>
> It is creating 8K hash table for each queue. Since we have 4000 queues,
Yes
> it used 32MB. I wonder why the current code needs 1024 hash buckets,
Hmm actually that's a leftover from when we played with bigger queue
sizes, I inadvertently forgot to change it back when pushing the rbtree
deadline update to Linus. It used to be 256. We can shrink this to 2^7
or 2^8 instead, which will then only eat 1-2K.
> when maximum requests are only 256. And also, since you are making
> request allocation dynamic, can you change this too ? Any issues here ?
No real issues to shrinking it, bigger problem if we move to larger
queues. With the rq-dyn-alloc patch, we can make the max number of
requests ceiling a lot higher and then the hash needs to be bigger too.
But for now, 256 entry should be a good default and suffice for the
future, I'll push that change.
-- Jens Axboe- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Mar 31 2003 - 22:00:27 EST