Re: Another Linux performance anomaly with 1+ GB files
Scott Laird (laird@pacificrim.net)
Wed, 12 Nov 1997 11:22:35 -0800
In message <346A018A.20FDB9D6@star.net>, Bill Hawes writes:
>Scott Laird wrote:
>
>> The pause also occurs when simply copying the same file from a 4k
>> filesystem to a 1k filesystem. I didn't try copying in the other
>> direction.
>
>Hi Scott,
>
>I'd hazard a guess that your problem is related to the different buffer
>size for the filesystem. Devices with buffer sizes different than the
>normal 1K will often have to allocate new buffers in refill_freelist, as
>they can't user the other sizes. In addition, since 4K buffers have
>only 1 buffer per page, it's much easier for the system to steal back
>the memory. If the 1 buffer on the page isn't in use at the time, the
>system can reclaim it. These factors when coupled with high memory
>demand could keep a task stuck in refill_freelist for a long time.
>
>Do the slowdowns occur when you're working with a normal 1K filesystem?
The only time I've seen the slowdowns, they've occurred during a read
from the 4k filesystem. I'm not sure if the WCHAN reported by ps is
accurate or not, but it was reporting 1261d9, which is in the middle
of try_to_free_buffers, according to my System.map.
As I reported in my first message, the system was using 64 MB of RAM
under 2.0.31pre10, but roughly half of that was buffers, and most of
the read activity has been occuring on the 4k filesystem.
Also, it's not really a slowdown -- it's a dead stop, at least for the
process involved. I can see spending a second or two looking for a
free page, but I've seen this take over an hour, all on a single
read(2), accoring to strace. It wasn't even a big read, just 8k.
Scott