Elias Torres wrote:
I work for IBM so I should be saying: yeah! let's throw more iron at the
problem and we'll be happier, but unfortunately more iron is not the
first thing we need here, good design/code is.
I agree, but I think that good design/code focused around cached dynamic
pages should meet most reasonable needs. (He says, carefully avoiding
the use of words like "any" or "unlimited"...)
I recently converted a project to use Servlet-based delivery of static
content that had been served with Apache httpd after years of avoiding
that for performance reasons. When we actually ran benchmarks, the
extra CPU used by the JVM vs. native Apache wasn't a real factor.
Are you serious? Wow! I can't believe someone would do such a thing.
I'll try to play along, was this Apache 1.3 or Apache 2.0?
It was Apache 2.0. I'm not saying that the specifics of our situation
can or should be applied to the general case. What I'm trying to say is
that you shouldn't assume that the conventional wisdom applies to your
situation. Doing some actual benchmarking often shows that what you
thought was the bottleneck wasn't. In our case we found that Tomcat was
able to serve the static content at over 200 Mbps when our deployment
configuration only used 100 Mbps Ethernet. Using Tomcat to deliver the
static content simplified SCM, deployment, authorization, and logging.
I wish we had run the tests earlier.
I'd love to hear from people who have large Roller deployments about how
they are configured and what the perceived (or better yet, measured)
bottlenecks are.
Wordpress is a single-site blog
There are multi-site versions of it like:
http://wordpress.com/
In fact, I believe there are (at least) two competing multi-site
versions of it. (One uses multiple DB's the other a single)
That's about all I know about it.
MovableType on the other hand has delivered the
scale, but they have static content. Dynamic features are great but need
to be carefully thought out and used sparingly if you want to play in
the big leagues.
Do you believe there is a significant difference between well-designed
and coded cached dynamic content versus generation of truly static files?
I think that we already have the potential of getting
into so much trouble with our template system that it will be impossible
to estimate the amount of hardware and memory we'll need to handle our
sites.
I'd be interested in hearing more about these issues.
We have little control over each blog's performance and how
it affects the rest of the site.
One last thing, I think we need to remember the Web 2.0 direction and
think about dynamic content differently other than with JSPs and
Velocity templates.
I agree. Which is one reason why I think we need view independence in
the blog UI layer.
-- Sean
