spin lock contention: lru_list lock

From: Manfred Spraul (manfreds@colorfullife.com)
Date: Sat Feb 19 2000 - 07:53:14 EST

I tried sgi's lockmeter with 2.3.46, and during a simple

        make -j4 dep

I get extremely long hold times for the lru_list lock:

sync_buffers(): up to 14 milliseconds, average 12 milliseconds. [called
10 times per second.]
__invalidate_buffers(): up to 8 milliseconds, average 7 milliseconds.[
called ~4.5 times/sec]

I assume that everything is cached:
* 192 MB memory.
* /proc/meminfo: Cached 122 MB
* slabinfo: dentry_cache: 17705 entries, inode cache 15174 entries,
buffer heads: 108189 entries.

Is it possible to optimize sync_buffers()/__invalidate_buffers()
further? I assume that with 1 GB memory, the hold times could become a
problem for a huge fileserver. 

--
	Manfred
P.S.: some good news: except do_close [up to 34 milliseconds], noone
else owns a spinlock for more than 3 milliseconds.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

This archive was generated by hypermail 2b29 : Wed Feb 23 2000 - 21:00:23 EST