Hardware question

[L.W.C. Nirosh]

I just got this question form a client and I cannot figure out a answer ..
can some one help me out here..



He need me to give the hardware configuration for a server that host a
system which does this...



An application that could potentially host 500,000 registered users that
would generate in excess of 5M Peronal Volume per hour.



How many processor do I need here (I know one processor can handle 12-16
parallel threads) I have done hardware estimation but this statement is not
clear to me.. just before going back to him thought to post it here..



Nirosh.

 

[Alberto Poblacion]

An application that could potentially host 500,000 registered users that
would generate in excess of 5M Peronal Volume per hour.

I assume that you mean a web server that is going to generate 5 million
pages per hour.

You have to do some benchmarking and find out how much time is needed on
average for each page that your aplication delivers. For instance, if you
find out that each page requieres an average of 100 milliseconds to
generate, then 5M pages will require 500 000 seconds. Since there are only
3600 seconds in one hour, that would mean that you need a large web farm
with at least 138 cpus. Since this is probably not acceptable, you would
have to optiize your application until the average cost for a pages goes
well below the original value, and you would then repeat this calculation to
find the number of cpus. This gives you a value that assumes that the page
requests are evenly distributed along time. Since there are likely to be
some times were demand will peak, you will have to provide some additional
processor capacity to cover the peaks.

How many processor do I need here (I know one processor can handle 12-16
parallel threads)

The default IIS configuration spawns up to 25 threads per processor
before starting to queue additional requests. This only means the number of
requests that shall be in the process of being computed simultaneously, but
it doesn't tell us anything about how fast they will be completed. That
depends on your application, which is why you will need to do some
benchmarking.

 

[Alun Harford]

I just got this question form a client and I cannot figure out a answer ..
can some one help me out here..



He need me to give the hardware configuration for a server that host a
system which does this...



An application that could potentially host 500,000 registered users that
would generate in excess of 5M Peronal Volume per hour.



How many processor do I need here (I know one processor can handle 12-16
parallel threads) I have done hardware estimation but this statement is not
clear to me.. just before going back to him thought to post it here..

You're far from clear.
Assuming you want to serve 5million pages per hour, you'd probably need
around 200 CPUs.

Alun Harford

 

[L.W.C. Nirosh]

thanks for the answer

Nirosh..

You're far from clear.
Assuming you want to serve 5million pages per hour, you'd probably need
around 200 CPUs.

Alun Harford

 

[L.W.C. Nirosh]

Thanks for the detail answer

Nirosh.

I assume that you mean a web server that is going to generate 5 million
pages per hour.

You have to do some benchmarking and find out how much time is needed
on average for each page that your aplication delivers. For instance, if
you find out that each page requieres an average of 100 milliseconds to
generate, then 5M pages will require 500 000 seconds. Since there are only
3600 seconds in one hour, that would mean that you need a large web farm
with at least 138 cpus. Since this is probably not acceptable, you would
have to optiize your application until the average cost for a pages goes
well below the original value, and you would then repeat this calculation
to find the number of cpus. This gives you a value that assumes that the
page requests are evenly distributed along time. Since there are likely to
be some times were demand will peak, you will have to provide some
additional processor capacity to cover the peaks.



The default IIS configuration spawns up to 25 threads per processor
before starting to queue additional requests. This only means the number
of requests that shall be in the process of being computed simultaneously,
but it doesn't tell us anything about how fast they will be completed.
That depends on your application, which is why you will need to do some
benchmarking.