NYCPHP Meetup

[David A.Roth]

Good question, Cliff. In addition to mere uptimes, I'm concerned about 
capacity planning. I don't feel comfortable recommending anything 
without being able to answer the real question of how much can it 
handle. This kind of reminds me of when I had the central air 
conditioner in my home replaced. Some installers would only do a Manual 
J Calculation to determine exactly what was needed, while others 
claimed that by the time they pulled up in the driveway they knew the 
number of tons size unit the home needed. I feel comfortable having a 
formula if one can be applied.

I wonder if anyone has taken servers in a lab environment and did 
benchmarks on it so that there were some sort of guidelines? That way, 
if marketing says "We expect X number of users accessing the web site 
at once..." then a nice formula to plug it into to help us gauge what 
CPU, RAM, bandwidth and number of servers that would be needed would be 
very helpful. This would vary depending on how a user is defined such 
as streaming video or submitting an inquiry form, for example.

As for determining the number of requests per second an Apache 
installation is capable of serving, there is a tool called 'ab' (Apache 
HTTP server benchmarking tool). I have some experience with this tool. 
I connected an older slow box on the LAN, and continued to increase 
traffic until it buckled. Of course, I wasn't able to perform such as 
test with a web hosting service because I'm sure it would be considered 
a Denial of Service attack, which is why I mentioned to do this in a 
lab environment LAN and then with permission to a live target web site.

For the specific web site, a load test tool could simulate X numbers of 
users doing Y activities simultaneously.

I'm thinking to gather all this test data, a package (open sources) 
could be installed on the source and target test machines and the 
results uploaded to a web site which would be a clearing house for the 
data. Anything like what I described (wished for:-) ) exist in open 
sources?

David Roth
rothmail at comcast.net

On Nov 16, 2006, at 5:43 PM, talk-request at lists.nyphp.org wrote:

> Date: Thu, 16 Nov 2006 14:28:27 -0500
> From: "Cliff Hirsch" <cliff at pinestream.com>
> Subject: [nycphp-talk] Bandwidth modeling -- extending the hosting
> 	question
> To: <talk at lists.nyphp.org>
> Message-ID: <002501c709b5$645dade0$12a8a8c0 at HirschLaptop>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Every service provider asks me, how much RAM do you want? How much
> bandwidth do you need?
>
> My standard answer is that on the first day, the application will have
> one user - me. After that, it's anyone's guess.
>
> Any thoughts on ways to estimate bandwidth requirements? Are there any
> modeling tools available?
>
> Cliff
> _______________________________
> Pinestream Communications, Inc.
> Publisher of Semiconductor Times & Telecom Trends
> 52 Pine Street, Weston, MA 02493 USA
> Tel: 781.647.8800, Fax: 781.647.8825
> http://www.pinestream.com <http://www.pinestream.com/>
>