Techniques for measuring impact of acceleration

Colin, any tips on other forums that might be able to provide this answer?

I agree with Colin, there isn't a forum to address this since we are talking more about performance testing, optimization and tuning.

 

 

To answer your question of what can one do to measure whether SSL acceleration will or will not impact a particular system, you have to pretty ask the following:

 

 

How many "new" concurrent SSL requests is your site generating per second?

 

How many "bulk" SSL connections is your site supporting per second?

 

 

The reason why these are really important questions is because you need to define how many new connections are coming in per second, which then determines the appropriate licensing model with the LTM appliance. The secondary question is useful because it will tell you which F5 model you should be using, I think the Bulk SSL on the 3400 is something like 75,000 or more per second, which is huge and quadruple that for the 6400/6800...it's some obnoxiously large number...so you get the picture.

 

 

Now, if you have a site that only needs something like 100 new TPS (trans per second) for SSL, you go with that license and you should be fine, the risk is if you ever bump up against that license then you start to impact your end-user populace (requests go in a wait state for a license to free up, and the resources on the F5 might start to go into overdrive as the proxy daemon has to keep track of all the requests and divy the licenses up).

 

 

To know what your site supports or can support is a different question probably for a mercury load runner forum, which is what I am very familiar with, in terms of baselining your application and trying to determine what your application can support.

 

 

Get the first 2 questions answered, that will tell you whether or not what you have will support your requirements, the secondary is only important if you have a high-volume SSL site, the first is very important.

 

 

-Wes