Re: new metadata reader/writer locks in integration-test

[Arne Jansen <sensille@gmx.net>]

On 21.07.2011 07:44, Arne Jansen wrote:
> On 20.07.2011 19:21, Chris Mason wrote:
>> Excerpts from Chris Mason's message of 2011-07-19 13:30:22 -0400:
>>> Hi everyone,
>>>
>>> I've pushed out a new integration-test branch, and it includes a new
>>> reader/writer locking scheme for the btree locks.
>>>
>>> We've seen a number of benchmarks dominated by contention on the root
>>> node lock.  This changes our locks into a simple reader/writer lock.
>>> They are based on mutexes so that we still take advantage of the mutex
>>> adaptive spins for write locks (rwsemaphores were much slower).
>>>
>>> I'm also sending the individual commits, please do take a look.
>>
>> Hi everyone,
>>
>> I just rebased Josef's enospc fixes into integration-test, it should fix
>> the warnings in extent-tree.c
>>
> 
> With the current integration-test branch I get very early enospc on
> a 7G volume create with -m single -d single and
> 
> fs_mark-3.3/fs_mark -d /mnt/fsm -D 512 -t 16 -n 4096 -s 51200 -L5 -S0 -R1
> 
> It enospces at about 20%, but I can continue to fill it up to 94%.

I tried to bisect this, but it turned out to be hard. Sooner or later
I get this early enospc on every revision, on some sooner, on others
later. At least the current for-linus branch is much worse than
integration-test.

> 
> -Arne
> --