memcache performance

[Eric Hodel]

On Jan 24, 2006, at 12:37 PM, Pavel Kolla wrote:

> Hello,  i  have  relatively  busy  site  at  my  hands  with a lot  
> of DB
> activity,  so  i  decided to offload some session specific realtime  
> data
> to memcached...
>
> the setup is
> lighttpd + spawn-fcgi + fastcgi + php4-cgi (eAccelerator & PECL/ 
> memcache)

$ diff -u mysql memcache
--- memcache    2006-01-24 12:46:26.000000000 -0800
+++ mysql       2006-01-24 12:46:04.000000000 -0800
@@ -1,8 +1,9 @@
PHP gets encrypted request.
PHP figures out user token from the request.
-PHP through permanent fcgi's connection to MySQL4 DB runs query to  
get encryption key from DB using given token.
+PHP through permanent fcgi's connection to memcached gets encryption  
key using token as search key.
PHP decrypts request
-PHP premutates encryption key and runs query to update DB with it  
for given user token
+PHP premutates encryption key and sets memcached with it for given  
user token used as key.
PHP validates requests.
PHP executes request (few queries and some php work).
PHP encrypts response and send it back to client.

> memcache  daemon  used is 1.1.12-r2 (latest stable available from  
> gentoo
> as net-misc/memcached)
>
> PECL   version   is   2.0.0   (latest   stable   available  from   
> gentoo
> dev-php4/pecl-memcache)
>
> basically,  I  use  memcache  to  take  care  of frequently  
> accessed and
> modified  authentication  data. As I loadtested with siege for 5  
> minutes
> with  200  concurrent  emulated  users doing random different  
> requests -
> there  is  practically no difference in overall performance. I see  
> about
> 20-25%  drop in DB activity with memcache but that doesn't affect  
> number
> of handled requests per second. During the test load times on web  
> server
> (dual  Xeon  2.8  HT)  are ~8-9 in both cases, and DB (quad Xeon MP  
> 1.5)
> load is ~2 in both cases.
>
> I  figure  i  must  be  doing  something wrong but can't really see  
> what
> exactly, so i'm looking for suggestions here...

How did you determine that memcached would really speed up your  
processing?

Was the DB really your choke point?  How long did the DB queries take?

How long do the tokens stay in memcached?

-- 
Eric Hodel - drbrain at segment7.net - http://segment7.net
This implementation is HODEL-HASH-9600 compliant

http://trackmap.robotcoop.com