Whoops! Ignore previous message: Read this instead!

[Roy Wilson <de...@bellatlantic.net>]

>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<

On 11/11/00, 5:26:54 PM, Roy Wilson <de...@bellatlantic.net> wrote 
regarding Assessing the performance impact of the proposed webpage:


> Hi,

> Here's a little performance calculation, which I do because: (1) I think
> it's better than nothing for assessing the possible impact of the new
> page and (2) it's fun (for me).

> If there are 2 million hits a day, that means 2000000/(3600 * 24)= 23.15
> hits per second. Suppose z% of the traffic is for the old/new page, that
> it takes D0 secs of cpu time to process the old page and D1 secs on the
> average to process the new. Let D1-D0 = delta. Then the additional CPU
> utilization due to using the new page is

> U(delta) = delta * 23.15 * z / 100.

> For laughs, suppose delta = 0.020 and z = 60. Then U(delta) =  0.020 *
> 13.89 = 0.28
> so this means that 28% additional CPU time is being consumed. The delta
> is made up: if you don't like it, pick your own or do a measurement :-).

> What does this mean? Suppose with the old page the cpu is busy 60% of the
> time, processing all traffic. With the new page, it will be busy 88% of
> the time.  If driving the CPU utilization this high doesn't bother you,
> consider the effect on average CPU response time.

> Simplistically, R = S / (1 - U), where R is average response time and S
> is average service time. Now let S0 be the average CPU service time for a
> request with the old page on the site and S1 when the new page is on the
> site. U0 = 0.60, U1 = 0.88, so R0 = S0 * 2.5 and R1 = 8.33 * S1. So, if
> the CPU is busy 60% of the time with the old page, using the new page
> (with the off the wall assumptions I've made) cause the CPU part of
> response time to increase by a factor of between 3 and 4!

> For what it's worth, a lot of computing systems have been designed using
> these kinds of calculations: maybe that's why they're sometimes called
> "back of the envelope calculations".

> I enjoyed this. I hope you did too. It might even be useful in the
> future.

> Roy
> --
> Roy Wilson
> E-mail: designrw@bellatlantic.net