General discussion

Locked

web server benchmarking

By wolfgangs ·
I am currently in a situation where I need to show to our board that our web servers are working at capacity. Can anyone point me towards throughput figures for web servers measured for different hardware/software platforms. Obviously, I would need rather recent data, i.e. benchmarks from before 2002 are not really of much value.
We are running a pair of DELL PowerEdge1850, 2GB RAM, 2x2.8 GHz CPU, Debian Linux (sarge), Apache 2.0, MySQL 4 (separate servers), so anything in this direction would be really appreciated.

This conversation is currently closed to new comments.

2 total posts (Page 1 of 1)  
| Thread display: Collapse - | Expand +

All Comments

Collapse -

2 things

by Jaqui In reply to web server benchmarking

pull a network throughput for the connections to the servers.

pull a cpu usage report from the servers.
every day for a month.

if the throughput on the nics is at 100% / the cpu usage is above 50% for any significant amount of time, then your server setup needs to be altered, through adding newer hardware or through adding more servers and balancing the load between them.

editing to add:

benchmarks for other machines aren't of any real use, the load your own machines are working under is the strongest arguement for the need to change them or not.

Collapse -

More figures

by wolfgangs In reply to 2 things

NIC throughput is not anywhere near 100%. These machines have gigabit copper ports and average max throughout (5 min. average) at peak times is around 7-8 MBit/s.
If cpu usage is the same as cpu load (as in : prcoesses waiting for the cpu) then we probably have a serious problem already. During peak times, cpu load is between 150 and 250 (with 2 hyperthreaded cpu's).
What happens though is that at peak times the servers start pumping. The page load time for the home page jumps up from around 0.8s to 2s and higher (sometimes up to 40s and more) and the load curve gets really bumpy. I would take that as a sign that the machine has no reserves left and whatever it can't process ends up in the queue (and that drives the cpu load).
Apache is configured for a maximum of 500 processes, but during peak times it only goes up until around 250-350.
what other figures could I look at ?

Back to Networks Forum
2 total posts (Page 1 of 1)  

Related Discussions

Related Forums