[SGVLUG] Inside Team Romney's whale of an IT meltdown

[Rae Yip]

Hm, it's not clear to me that's a panacea.

While a caching proxy (which may be integrated into some software load
balancers) will certainly reduce the load for static content, if the
request involves dynamically generated data, I still think the tier
behind the LB will see more load when there are transmission delays -
especially in a multiple round-trip interaction (more common with Web
2.0 and phone apps).

I guess it would definitely help with the SSL connection overhead,
from the web server's point of view. But you now have to worry about
the load of the LB, since it becomes the bottleneck; while these
devices usually have a higher vertical limit, the point is you still
want to do capacity planning with real-world network delays in mind.

In any case, if Romney was using a single web server and single app
server as alleged, there probably wasn't an LB.

-Rae.

On Sun, Nov 11, 2012 at 11:44 AM, Christopher Smith <cbsmith at gmail.com> wrote:
> On Sun, Nov 11, 2012 at 11:31 AM, Rae Yip <rae.yip at gmail.com> wrote:
>>
>> A common issue when you use benchmarking software to stress test a web
>> app, is that you aren't simulating the connection latency that HTTP
>> over actual Internet imposes on a system. This can have a big
>> difference on the queueing behavior (ie. load) of the system.
>
>
> Yeah, except a load balancer tends to eliminate that problem.
>
> --
> Chris