[Asterisk-Users] 2 4-port T1 cards

[Steven Critchfield]

On Wed, 2003-05-28 at 14:11, Ryan Butler wrote:
> On Wed, 2003-05-28 at 13:30, Steven Critchfield wrote:
> 
> > Don't use Mysql. if you ever have had to deal with it in a production
> > environment that works it over, you will know that as it reaches it's
> > limits, it starts a death spiral that is very difficult to recover from.
> > For our software on a dual P3 866 with a gig of ram, the limit was
> > around 1.5 queries a second fairly mixed update, inserts, and selects.
> > Total file size of the database was under 200meg, and was fully cached
> > so even though we had hardware raid 5 across 4 10K rpm ultra160 drives,
> > it shouldn't have mattered for the selects.
> > 
> 
> I'd check what configuration problem you have with mysql then.
> 
> Uptime:                 48 days 1 hour 42 min 6 sec
>  
> Threads: 15  Questions: 376005206  Slow queries: 22  Opens: 214  Flush
> tables: 1  Open tables: 64 Queries per second avg: 90.531
> 
> 
> This is a mixture of inserts updates and selects as well.

This is partly why I qualified it with the statement "with our
application". I admit there was several things I inherited that are way
sub optimal. Mysql didn't help matters when it went into it's death
spiral though. Not all of it was the fault of Mysql, it just doesn't
handle it gracefully. My biggest grip beyond table level locking was
busy idling a process while the table was locked instead of giving up
the processor so it could service the thread that had the table locked.
I'd end up with 5+ processes queued up waiting for a table to be free
using as much CPU as they could blocking on the table.   

> slashdot.org runs mysql at loads way over this, and it continues to run,
> although its very select heavy most likely.

Actually it is mostly cached. For the case of slashdot, 95% or more only
read the front page, and the values of messages do not need to be
accurate. Of the 5% that read deeper, very few actually post, and only
the posts require anything more than selects.
-- 
Steven Critchfield  <critch at basesys.com>