Did you set a ttl for the items? How similar to your real data is this
test? It seems to be a very academic test and may not reflect your
actual usage even if you did see the results you hoped for. I would
suggest easing into memcached, watching the hit/miss ratio with cacti or
some other monitoring and then deciding if memcached is right for you.
I can tell you that the LRU works fine for me. I have never had an
issue with things going away when I was not expecting them to. And as
always, it is a cache, not a storage system. Don't put stuff in there
that can not be recreated easily.
Mathieu Viel wrote:
Hi guys. I have a question about the cache structure.
I want to use memcached to store some frequently used data chuncks and
the idea is to always have in memcached the most frequent requested
data; if i insert some new data when the memory for memcached is fully
used, i want the least requested data to be replaced.
I noticed in the FAQ that the cache structure is an LRU so i thought
"great i'll have almost nothing to do to implement this!" but, in order
to make sure, i ran some tests and i am dissapointed about the results.
With a 1Mb memcached, i first set an item with "1_1_1" as key and "1" as
value; i then get the value of this item (which is working ^^) and then
set some new randoms items i will never retrieve in order to fill the
memory (i made sure the key value will never be "1_1_1" for these
items); i then get the value of my "1_1_1" item to check if it's still
there. This "simple" process is working fine but when the memory is
filled, after some time, i can't get the value of my "1_1_1" item
anymore :S
I tried to check in verbose mode (-vv) and i can't get any interesting
information about this "issue".
I thought that since i frequently get the value of this item, it would
not make it the LRU one so its would never be somehow erased. Am i
wrong? I know about the different slabs sizes but since i always set the
value to "1", i think all these items are in the same category.
Thanks for help!
--
Vivi
--
Brian Moon
Senior Developer
------------------------------
When you care enough to spend the very least.
http://dealnews.com/
