[thelist] Website speed smart practices

[Benjamin Hawkes-Lewis]

On 18/1/09 20:29, Bob Meetin wrote:
> I tried many of the things Benjamin noted, but was not getting a lot of
> success in consistent improved performance.  The YSlow stats talked
> improvement, but my eyes said no.

It's perhaps worth stressing that YSlow is primarily a frontend 
performance diagnostic tool, rather than a performance measurement tool. 
In particular:

1. The grades it provides (e.g. "A (92)") are based on matching a  set 
of criterion based on frontend performance best practices not a direct 
measure of performance at all. Obviously, the closer you match the best 
practices, the higher the second rating will be. But other factors can 
totally blow away the effects of these practices - e.g. networking 
problems, giant images, etc. Equally, there are also special 
circumstances where what's generally best practice simply isn't appropriate.

2. To exclude the effects of network latency, you'd want to aggregate 
multiple requests (ideally from multiple locales) to produce any 
performance figure, especially when trying to isolate the effects of 
small changes.

> I tried a couple different options with ETags, but other than a ySlow
> grade improvement I could see no noticeable load speed improvement.

Note that:

1. ETags make no difference to an uncached request.

2. If you host all your content on one server, the problems with ETags 
are limited to minor increases to header size ( 
http://developer.yahoo.com/performance/rules.html#etags ).

> For troubleshooting problems caused by network or webserver host what
> might be recommended besides 'ping' and 'traceroute' for a Linux desktop?

SSH access to a third machine on a different network would probably be 
handy. ;)

--
Benjamin Hawkes-Lewis