[OpenAFS-devel] good and bad performance of memory cache
Edward Moy
emoy@apple.com
Thu, 6 Feb 2003 15:03:12 -0800
On Thursday, February 6, 2003, at 01:27 PM, Lyle Seaman wrote:
>> On the server side, memcache resulted in about 1.4 million input and
>> 0.92 million output packets, while diskcache was 0.80 million input
>> and
>> 0.90 million output. For the fileserver process, 1:46 cpu minutes for
>> memcache and 1:03 cpu minutes for diskcache. The 1 minute load
>> average
>> never exceeded 0.45 in my samples, though the 5 minute load average
>> peaked at 0.28 for memcache, 0.21 for diskcache.
>
> neat. So, what do you think is happening?
>
>> NFS ... resulted in 0.87 million input and 0.62 million output
>> packets.
>
> That would be as seen from the server, right?
Yes, from the server side.
The results of iozone itself might provides some clues (I split the
results into two parts; units are Kbytes/sec):
random
random
KB reclen write rewrite read reread read
write
memcache 131072 64 4223 1262 2328 2323 206218
3810
diskcache 131072 64 4114 1821 5849 5847 157343
3599
bkwd record stride
read rewrite read fwrite frewrite fread freread
memcache 5260 211525 2891 3603 1267 2288 2284
diskcache 11370 87494 7071 4146 1816 5865 5852
Notice that except for random reads (where the memory cache is showing
its advantage), all other memcache reads are less than half as much the
corresponding value for diskcache.
Also, the rewrite speed is a lot slower than the first write. This
also happens for file sizes smaller than the cache size, at the smaller
record sizes, but as the record size increase, the rewrite eventually
becomes larger than the first write. But once the file size is larger
than the cache size, the rewrite speed tends to level off at a low
value, while the first write value is still many times higher.
And read values for memcache are also less than memcache write speeds
(except the rewrite value is worse than reads).
------------------------------------------------------------------------
--
Edward Moy
Apple Computer, Inc.
emoy@apple.com
(This message is from me as a reader of this list, and not a statement
from Apple.)