Forecasting BW and impact on speed

Better be ready than sorrry!

[henry0]

Prerequisite:
Minimum 1 M members
Minimum visits 30M/Year

How to measure the impact of PHP generated stats bar
On about 100 items/10 per page
Any page could be viewed at least 15000 times /day
So during peak hours as many as 2000 similar pages/hour could be requested
This will barely represent 10% of the other offered services. (although other services won't call for huge BW needs)

If as expected it reaches that level I will ahead of time split servers load
So here we only are speaking about web server.

[jatar_k]

what's BW?

<added>ah, wait, bandwidth

I don't quite understand what your question is Henry

[henry0]

Well, my question is generic.
Within the described environment
Having that many statistic bars (10 pages each with 10 bars) graphically depicting a percentage and PHP image generated.

Since each page could be requested a minimum of 2000 times/hour
how ahead of time could I be able to define the server power requested.

If it could drastically slow down operation I could envision instead of a bar some form of star system based on percentage, but it will be less graphically correct

[jatar_k]

ok, sorry, now I get it

I think any image generation could significantly affect the performance of any page

why not benchmark it?

[henry0]

Thanks, I could indeed benchmark the performance

but isn't it going to be only reflecting a given environment (the one where the test is performed)

What I try to define is: If I want to make it working as defined could I in return forecast a server config that will serve it stil in an acceptable time manner.

[jatar_k]

I'm not sure about forecasting an actual config

certain server specs will increase performance in various ways but it is almost impossible to figure out what config will be optimal for your setup. Though one thing is always true, if you have an issue, throw more RAM/servers/cpu power at it and it will speed up (to a point).

I have to say you are looking at it the wrong way, imo.

If you design something to be server intensive then you are going to run into problems at some point. If you design something to run light and tight then when you do have problems you are going to have more than enough traffic to warrant a change in server setup.

optimize the software first, worry about servers after. ;)

A dressed up pig is still just a pig

[henry0]

OK received loud and clear;

Will start with a decent machine, benchmark img php generated and see where it goes.

How do you create traffic test, is there a way to mimic intensive usage
asides hiring an army of clickers :)

[edit]
maybe testing traffic should be another thread?
[/edit]

[edited by: henry0 at 9:57 pm (utc) on Nov. 1, 2006]

[jatar_k]

there is software around that does it, I've used some for 'load testing' but have no idea what it was called.

[StupidScript]

Load balancing servers should be able to distribute the load evenly even if it jumps in an unpredictable way.

The BW question might be answered only by real-world useage, as different services might be used more at some times than at other times, although some trend analysis could help predict near-term needs.

Am I out to lunch on my interpretation of the question?

[henry0]

Not at all, you are right on the spot.
Load balancing is part of the setup,
as I said trying to foresee usage is the toughest part
much more due to users' editable content generated.

I would like to be able to test it with as many requests and queries as possible.

Remember what happened (+/- 2 months ago) to a poorly designed FR Gov site, it was not able to support the load only a couple of days after the "D" day.

[StupidScript]

As jatar_k mentioned, there are lots of load testing tools freely (and expensively) available. Google "website load testing" for a bunch. You can also find a few here [softwareqatest.com].