[time-nuts] Sparkfun

[Lux, Jim (337C)]

On 1/7/10 10:14 AM, "John Miles" <jmiles at pop.net> wrote:

> A couple of thoughts --
> 
> - Refusing to let people into the site at all is probably the best approach,
> if you find yourself completely swamped with traffic.  A customer   who sees
> 10-second ping times is likely to be even more annoyed than one who gets a
> 404.  What they should have done, though, is served those people a static
> text page apologizing for the delay and inviting them to try again later...
> while at the same time trying to discourage them from hitting F5 over and
> over.
> 
> - If their hosting provider isn't capable of temporarily boosting their
> bandwidth as needed, that's kind of lame.  It should be almost impossible to
> "slashdot" any professionally-run e-commerce site these days.  If nothing
> else, they could have rented some time on EC2 to test their connectivity
> beforehand.
> 

My wife deals with this sort of thing (supporting new customer facing
websites) at a Fortune 500 company, and they still get bitten
occasionally... You can do all the planning, arrange for fallbacks, and
still stuff happens.

It's tricky to serve alternate static pages, especially since you probably
have a multi tier architecture to achieve persistence (after all, each HTML
"GET" conceivably goes to a different server thread, so you need some clever
way to keep track of session state for all the users.).. If you had a single
threaded server, there's no way to serve an alternate page if you get backed
up, and if you have a multithreaded server, it's tough to keep track of the
state.  A lot of folks just tough it out and hope they have enough capacity
to make it through.  You do a cost benefit analysis... "do I spend 50K on
labor to implement a bullet proof solution, or do I just fling it out there
and hope for the best"


 It's if it happens over and over at the same place, I get cranky, but this
probably isn't in that category.