Re: [PATCH] fix IO hangs

[Nick Piggin <piggin@cyberone.com.au>]

Martin Schlemmer wrote:

>On Fri, 2003-09-05 at 19:04, Jens Axboe wrote:
>
>>On Fri, Sep 05 2003, Jens Axboe wrote:
>>
>>>>Jens, if insert_here is dead, there is no point to passing back a hint
>>>>because it can't get back to the elevator anyway.
>>>>
>>>>I'd very much like to kill insert_here and be done with it. If another
>>>>io scheduler comes along with a good use for it then the writers can
>>>>come up with an elegant solution ;) Hey, if they know a NO_MERGE return
>>>>means an insert will soon happen under the same lock, they could keep
>>>>it cached privately.
>>>>
>>>Agree, lets just kill it off.
>>>
>>Here's the patch that kills it and its associated logic off completely.
>>Nick, I'm not too sure what the logic was for stopping anticipation in
>>as_insert_request() (the insert_here tests were somewhat convoluted :),
>>I've added what I think makes most sense: stop anticipating if someone
>>puts a request at the head of the dispatch list.
>>
>>It also makes the *_insert_request strategies much easier to follow,
>>imo.
>>
>>
>
>Hmm, do not know if its just me, but I just got two processes (cp's)
>in D state.  They did complete though, but throughput was not good.
>Any tips on getting it debugged ?
>

If they complete at all, then its very unlikely for them to be slowed
down (AS can do this a bit though). If you could get a comparison with
my patch that started the problems backed out it might tell me
something. Thanks.