Preparing the site for habraeffect

Question

Nginx

I am preparing the site of my small candle factory for the expected increase in the number of visitors due to a couple of events. I made a bunch of apache2 + nginx, set up the distribution of statics, compression of styles, other differences, in some places scrolled through the optimization.

Apache runs in the prefork:

StartServers 2
MinSpareServers 3
MaxSpareServers 4
MaxClients 5
MaxRequestsPerChild 3000

Ran ab - about 1700 rps, drove LoadImpact - delay about 2 sec with 50 clients, 1.2 s with 30 and 40 clients. The average stone load on 30-40 clients is about 35%, memory is within 280-290 MB for the entire machine.

Axis - Squeeze, stone - two cores from Xeon E5420 (I'm sitting on a VPS), RAM - gig.
What could I have missed, and which way to dig? Is it worth loading more RAM to unload the stone, or is it easier to move to a normal server with a planned start of 3k uniks (10-15 views per piece)?

Answer 1

On a slightly better hardware configuration is:
StartServers 50
MinSpareServers 50
MaxSpareServers 200
MaxClients 256
MaxRequestsPerChild 500
RAM eats up almost a gig at minimal loads. But 3-4 SpareServers are clearly not enough

Answer 2

What kind of influx of visitors do you expect?
If we are talking about a habraeffect or a similar amount of traffic, it is useless to turn Apache settings on a VPS with a gig of memory. Will blow away with hosting.

Answer 3

I have shared hosting, twice Habraeffect without delays and problems - 10,000 uniques per day + 3-4 views per each (on a peak day). What am I doing wrong?)

Answer 4

To look at the site - if it is not too dynamic or you can simply redo it - separate the dynamics and, as it were, static parts of the pages - then you can make the site give as much as the channel allows. The main thing is to skip as few requests to apache as possible, process nginx to the maximum.

Answer 5

you can also cache blocks in nginx or varnish, and if you have enough funds - Akamai :)