The Art of Computer Systems Performance Analysis
-
Jonathan
- Grand Pooh-Bah
- Posts: 6722
- Joined: Tue Sep 19, 2006 8:45 pm
- Location: Portland, OR
- Contact:
It is easy to read.
Couple points of interest from dipping into parts here and there yesterday. First, the book covers simulation, analytical models, and measurements with equal weight to each. I personally spend so much of my time doing simulation that I forget about producing analytical models and even measurements. Second, it doesn't touch algorithms or O() analysis. It's a down and dirty transactions per second kind of book. Third, it has a treatment of the games people play when reporting perf which is pretty neat. There's also some best practices about producing graphs in there.
Parts 1 & 2 didn't seem too important to me. After all, I do perf analysis for a living. These parts may be better if you don't worry about perf all the time, perhaps.
Parts 3 and 4 are killer and the reason I wanted to borrow the book. Part 3 is the statistics of perf analysis and something my dinky 200 level stats class did not touch on. Linear regressions are particularly interesting, but I'm also going to dig into the bits about reporting means properly. I was recently forwarded a ppt by a famous someone in ECE on this subject; I'll see if the material exists on the web and post a link here. Part 3 I'm going to have to study. Part 4 is experimental design. The bits about fractional factorial design are interesting in the extreme, as I do this type of analysis all the freakin time.
Part 5 is about simulation. I haven't really looked into this part yet, but I suspect there's not much in the book that isn't already part of the tribal knowledge in my office. Best practices already incorporated into our way of doing things, that type of thing. If I were creating a model from scratch I'd be more interested.
Part 6 is about queuing theory, which I guess was a bigger deal ten years ago? Seems like a good way to analyze a system's perf if you have a very short time and need to make a business decision now, but I don't anticipate using it in my work.
There's also some math tables and street addresses to IEEE and ACM and so forth. It is enchantingly pre-WWW.
I think it is worth having.
Couple points of interest from dipping into parts here and there yesterday. First, the book covers simulation, analytical models, and measurements with equal weight to each. I personally spend so much of my time doing simulation that I forget about producing analytical models and even measurements. Second, it doesn't touch algorithms or O() analysis. It's a down and dirty transactions per second kind of book. Third, it has a treatment of the games people play when reporting perf which is pretty neat. There's also some best practices about producing graphs in there.
Parts 1 & 2 didn't seem too important to me. After all, I do perf analysis for a living. These parts may be better if you don't worry about perf all the time, perhaps.
Parts 3 and 4 are killer and the reason I wanted to borrow the book. Part 3 is the statistics of perf analysis and something my dinky 200 level stats class did not touch on. Linear regressions are particularly interesting, but I'm also going to dig into the bits about reporting means properly. I was recently forwarded a ppt by a famous someone in ECE on this subject; I'll see if the material exists on the web and post a link here. Part 3 I'm going to have to study. Part 4 is experimental design. The bits about fractional factorial design are interesting in the extreme, as I do this type of analysis all the freakin time.
Part 5 is about simulation. I haven't really looked into this part yet, but I suspect there's not much in the book that isn't already part of the tribal knowledge in my office. Best practices already incorporated into our way of doing things, that type of thing. If I were creating a model from scratch I'd be more interested.
Part 6 is about queuing theory, which I guess was a bigger deal ten years ago? Seems like a good way to analyze a system's perf if you have a very short time and need to make a business decision now, but I don't anticipate using it in my work.
There's also some math tables and street addresses to IEEE and ACM and so forth. It is enchantingly pre-WWW.
I think it is worth having.
-
Jonathan
- Grand Pooh-Bah
- Posts: 6722
- Joined: Tue Sep 19, 2006 8:45 pm
- Location: Portland, OR
- Contact:
Summarizing Performance Is No Mean Feat seems to be the abstract, I can't find anything else on the web.