Re: [cobalt-developers] Concurrent Users

["Steve Werby" <steve-lists@xxxxxxxxxxxx>]

"Jörg Jan Münter" <support@xxxxxxxxxxx> wrote:
> does anybody have an idea how many current web-users a RaQ4r with 512
MBytes
> may do? Lets say it is a web-page half html and half PHP with MySQL, the
> MySQL-Database is running on a different machine. Could more than 100  or
up
> to 200 Users simultaneously be possible? I talk about "real" visitors who
> spend some time on every page

If I understand you right, you're asking if 200 people can be visiting the
website you describe at the same time.  Absolutely.  I expect you could
handle thousands of concurrent users, though it really depends on how much
CPU time each page requires to load, how long each user stays on the page
before visitng another on the server and what else is happening on the
server at the same time.  I have built database-driven, content/data-heavy
sites for clients hosted on RaQ servers that have served up pages to several
thousand concurrent users.  For argument's sake, let's say that the average
user spends 20 seconds on a pertinent content-heavy page and 3 seconds on a
non-pertinent content-light page.  Since the average HTML page takes a small
fraction of a second of CPU time and the average PHP time takes a small (but
relatively larger) fraction of a second of CPU time, assuming page requests
are somewhat normally distributed you'd be able to accomodate "concurrent
users" to an order of several magnitudes greater than "concurrent processes"
which is probably the true bottleneck in your scenario.

> , i don't mean those extreme tests without
> waitcycles between jumping to the next page.

I understand what you're saying, but it doesn't hurt to do some testing to
see.  You can always use a benchmarking program like ab (ab is on your
server already, do "man ab" for more info.) or find/write a simple shell
script that mimicks real expected arrival and usage patterns.  Your users
probably arrive randomly and are distributed according to something like a
Poisson probability distribution and the amount of time they stay on a page
before moving onto the next page is probably also random and normally
distributed.  It's probably overkill to model something like this, but you
could always assume a constant X users using the site, pick a fast-loading
HTML page, a light PHP page and a heavy PHP page, assign a probability of
visiting each so they total to 100% and assign a corresponding average time
on each page and normal distribution and fire up the simulation.  Then you
just repeat each simulation a few times, record whatever criteria you want
to record, keep increasing the number of concurrent users and repeat to find
out what your system can really handle.

If you search the internet, you can probably even find scripts that do
pretty close to what I describe.  I've done such simulations myself, but I
don't have any scripts handy that will run on your RaQ.

> Or by any chance do some official notes of Sun Cobalt exist concerning
those
> questions about performance?

Just their marketing text that states (or used to) how many hits per day and
emails per day a RaQ could handle.  Of course, they never revealed how they
arrived at their numbers and such figures wouldn't be useful to you even if
Cobalt did reveal what assumptions they made.

--
Steve Werby
President, Befriend Internet Services LLC
http://www.befriend.com/