I was thinking that for Apache the received wisdom is around 50 concurrent users per GB of RAM. I assume that figure will improve by using lighttpd, but does anybody have an idea how much?
Any thoughts appreciated
Hi Howard,
Good experience with lighttpd here. We are running a server with 30 separated instances of moodle - the profit of cache mechanisms should be much lower compared to a single instance.
Result: 30% more available RAM depending of the settings of the PHP-Cache s (I use all of freed RAM to cache MySQL). Apache is much - on our server - slower with identical configuration (fastCGI) and even with mod_php (with lighty you must use PHP in fastCGI-mode - which is much easier in configuring it safe).
Further apache in general is harder to configure, requires a new start after adding vhosts - lighty not (mod_evhost) and you have to deal with all these context switches - with lighty there is only one process handling all requests (a totally different concept to apache).
If you have money - buy big hardware and use the well documented standard (apache). If not, give lighty a try.
regards,
Maik
Well, I have money to buy big hardware. But I need very big performance. I was hoping that lighttpd might provide some insurance.
Some more details:
Server:
3500 Athlon XP
2GB RAM
S-ATA-HDD (non raid)
state information:
netstat (open tcp-connections, not waiting): 40 (5 minute average)
RAM-consumption: 600MB
Load: 0.8-1.2
usage:
Server:
3500 Athlon XP
2GB RAM
S-ATA-HDD (non raid)
state information:
netstat (open tcp-connections, not waiting): 40 (5 minute average)
RAM-consumption: 600MB
Load: 0.8-1.2
usage:
Users not using quiz or chat module. If this happens load will increase up to 8-10 - but he'll deliver still pages in 4s-5s (much faster than moodle.org the last days...).
Server stays stable in any case - just becoming slower. In the described situation fastCGI-processes were waiting for mysql (so load had to increase though RAM and CPU were fine...). So it was my fault and not lighty's.
I just tuned MySQL some more (extremly increased caching space) - so I have no more representive data for the moment including quiz / chat.
Server stays stable in any case - just becoming slower. In the described situation fastCGI-processes were waiting for mysql (so load had to increase though RAM and CPU were fine...). So it was my fault and not lighty's.
I just tuned MySQL some more (extremly increased caching space) - so I have no more representive data for the moment including quiz / chat.
That's lighty combined with squid in reverse proxy mode (cache hitrate 70-85%).
I figured out that I'll have to keep away my users from the backend
... Squid helps much and gave much benefit and insurance.
Maik
I figured out that I'll have to keep away my users from the backend
Maik
How are using netstat to get that figure?
I'm a bit hazy on the best way to get concurrency figures.
I'm a bit hazy on the best way to get concurrency figures.
The package munin helps with all of this. Nice graphs are created using rrdtool. You can do it with nagios either - canons on small birds...
Maik
Maik
Ahh thanks - I'll take a look at that 