Jim Panttaja

Yahoo has a web site on Exceptional Performance and Steve Souder, Chief Performance Yahoo! has recently authored a book (O’Reilly) called High Performance Web Sites.

Yahoo also makes available a tool, YSlow, that helps you to assess how you measure up against the Yahoo guidelines.

The advice is clearly focused on sites like Yahoo – and as a number of blogs have suggested, you need to take the advice with a grain of salt. But it is a good starting list – especially if you look at some of the follow on blog entries (many of which are referenced from the Yahoo site).

Phil Haack’s blog describes his experience using Coral Distribution Network (it slowed down his site).

Jeff Atwood’s blog entry in Coding Horror is called YSlow: Yahoo’s Problems Are Not Your Problems.

Look especially at the follow on discussions on Jeff Atwood’s blog.

I now live in two places – one with a 40 megabit connection, the other is much slower (and I am now in my car – and seeing about 800 kbps). It is worth making sure that your engineers are not always running on the the high speed connection – with the servers next door. Sometimes performance problems are masked by really high speed links, or small development databases. You need to simulate the real world when you are evaluating your system. And it may be worth simulating something even worse than the real world. Several years ago, I was running an application that was developed with a development environment. I was trying it on a dial-up line (remember those), and noticed that the application always repainted the screen twice. It was a serious bug in the tool that no one had ever noticed – because on high speed networks, you couldn’t see the repaint.

Use other people’s advice (with a grain of salt), and do your own ‘real world’ evaluations.

Update:

A couple of my colleagues pointed me at some tools to facilitate load testing:

• JMeter
• Grinder
• Netem

Update:

The comments keep coming in. There is a post today on O’Reilly Radar  pointing to A Great Performance and Operations Blog called High Scalability by Todd Hoff. Lot’s of posts about Amazon, Twitter, Hadoop, etc.

And my friend and former colleague Jeff Dao comments:

Some of the suggestions I know of, and some I don’t – like get servers closer to the users, and order you load scripts matter, respectively. Bottom line is, distance and data size matter.

When thinking about this, I always think of water delivery system – pipe size, water volume and water pressure (bandwidth, data size and latency).