Mailinglist Archive: opensuse-web (64 mails)

< Previous Next >
Re: [opensuse-web] Please update wiki staging server
I will see if I can keep working with Scott on this today. It's
quite frustrating, because there is no conceivable reason for it to
be slow... except that it is :-)

Traffic stats.
Average Average Load Time for
Slow (2.326 Seconds), 73% of sites are faster

We were actually talking about the stage server, which was taking something
like 30 seconds to load. Turns out that the memcached server had been turned
off, so it was spending all that time to query something that wasn't running.
We flipped it back on, and it's all fixed now!

As for production, some of you may remember me talking about this a year or two
ago. When I took over the wiki, I made a ton of performance enhancements. We
now use almost every type of caching available (opcode, memcached, file
caching, and an accelerator). I also did some Apache performance tuning, which
was desperately needed. The wiki is much faster than it used to be, even as we
have added a bunch of new features.

I also mentioned that we still have a long way to go, but most of it should now
be front end optimization. Things like combining javascript files, using
sprites where it makes sense, and removing serializations as much as possible
will give us the biggest wins at this point. If we can go from 40+ resources
down to under 30, and get them parallelized and ordered as well as possible, we
should expect to cut that loading time by at least a second.

In addition to the front end optimizations, I have noticed that I'm not using
gzip and browser caching as much as I could. I'm looking into that right now.
It should help us out a bit, but not nearly as much as what I mentioned above.

Sometimes it is stuck at loading (nothing) and then after repeating
request it is all loaded in a very short time.

I have noticed that too. Sometimes it gets stuck on the main page, sometimes
on counter, and sometimes on beans. Just now, I noticed beans taking something
like 30 seconds to come back. I can't speak for counter, beans, and static, as
those are hosted elsewhere, but I have looked into why sometimes the main page
doesn't come back. So far, the main culprit appears to be our ancient load
balancer. The good news is that we are already looking at replacing it.

Speaking of that, we are looking at some load balancer options that can do a
lot for us. One vendor can implement SPDY and is capable of downsampling
images on the fly for mobile devices. Another is capable of TLS false start,
which can cut SSL negotiation time by 25% or more. All of them will give us
TCP optimizations, TLS 1.1/1.2, and better IPv6 support, so good things are
coming down the road. Stay tuned :)

< Previous Next >
Follow Ups