Interesting (mysql?) problem 2006-02-15 - By Ken Snider
Back
So, I've come across an anomaly.
I have a mysql database running on 64-bit 4WS. The database handles about
100-150 non-persistent connections per second from a few hosts over the
network. The server is started with skip-name-resolve, and all hosts use ip
addresses to connect. back_locg is set to 1000, and max_connections is set
to 1000. max_user_connections is 0. max_used_connections is only around 200,
as well.
The database being used is tiny, and the selects that are read come almost
entirely from the query cache. In short, this box does *almost* no I/O.
At any given point, the CPU is about 10% utilized, according to vmstat.
The network connections of all nodes have been tested with ttcp and can push
95+Megabits easily. Duplex has been checked, etc.
However, for some reason, with mysql is under load, interactive response on
that server becomes "choppy", that is, pauses of a few seconds here and
there can be seen in connections. Also, a connection to the mysql server is
usually subsecond, however every 5-10 seconds connections will take 3-4
seconds to establish, for no reason I can make out.
In short, this box is bored - it's doing nearly nothing and I cannot find a
resource contention issue of any sort - yet it would appear that, for
whatever reason, delays keep popping up on connect. I'm beginning to think
I'm hitting a TCP stack limitation of some sort. Any ideas?
--
Ken Snider
--
Taroon-list mailing list
Taroon-list@(protected)
https://www.redhat.com/mailman/listinfo/taroon-list
|
|
