5 ms·

CS193H: High Performance Web Sites

112 points by AbyCodes 15y ago

riffraff 15y ago
did anyone else just assume by default this was cs193h-class.org ?
- blcArmadillo 15y ago
  haha. yes.
  - worldimperator 15y ago
    I guess after the first three weeks of ai-class.com, judging from the several stability adventures, the makers of the site could offer the knowledge of that course as well ;-)
cleverjake 15y ago
Steve Sounders is a wealth of information. His http archive (http://httparchive.org/ http://httparchive.org/) is already showing fundamental changes in design, as well as regularly giving good tips on high-scalability sites. I am jealous of those who get to take the full course. =]
wunderfool 15y ago
losing some respect for stanford's vaunted CS program here...this is devry territory. i would expect material for a stanford degree to be a bit more rigorous than regurgitating the recommendations of yslow.
prophetjohn 15y ago
I know it's not technically computer science and it's more vocational, but I wish my school offered classes like these. Something like this would be more beneficial to me than, say, Theory of Automata, I think.
- lightcatcher 15y ago
  really? I'm actually glad my school doesn't offer vocational classes like this. Optimizing a website on the client side is the sort of thing you read a blog post for and then go do. I don't enjoy it at all when I actually have to do it, I certainly wouldn't want to take a course where I have to do this.
  - beagledude 15y ago
    that may be the newbest thing I've read on hacker news. It only shows you haven't worked in high performance environments.
    - lightcatcher 15y ago
      Do you believe that there aren't more useful things to learn in school? Stuff like this is fairly trivial to teach to yourself. I believe the time would be better served by taking a class in something like game theory or any one of the other mind-expanding courses offered (especially at a school like Stanford). As for the actual content of the class, all you really need to tell someone is that various things can make pages load slowly, and then link them to some guide such as this: http://developer.yahoo.com/performance/rules.html http://developer.yahoo.com/performance/rules.html . Also, I still have this opinion after working last summer in what I would consider a high performance environment at a YC company.
      - beagledude 15y ago
        I believe that CS courses should include more real world practices and situations. After interviewing hundreds of cs grads over the past decade it only re-enforces that belief.
      - simonw 15y ago
        I actually think high performance web architecture could make for an excellent mind-broadening course. Consider the things you need a deep understanding of: - DNS, how it works, DNS level load balancing, DNS client implementations in operating systems - Internet routing architecture and CDNs - Browser internals: how modern browsers access the network, how their parsers work, how they parallelise requests - In-depth understanding of both JavaScript and the DOM - How CSS style engines work - Different image file formats - How compression works - How to introspect and monitor web page performance - How to reverse-engineer complex web sites
mmaunder 15y ago
This isn't bad, but it's client focused. Perhaps there should be a second module that focuses on specific web server architectures (apache vs nginx and lighttpd), reverse proxying performance, performance of web servers talking to different data stores and the performance of those stores. Also server-side caching architectures as well as server hardware and what works best for web servers, caching servers and data stores.
- Mikushi 15y ago
  Agreed on that. I am actually currently working on a book (PHP oriented) on high performance / traffic website architecture. Worked 4 years at a Top50 Alexa and been eager to share my knowledge (was forbidden to talk publicly about most of it while i was working there).
  - mmaunder 15y ago
    Sounds great, I'd love to learn more so email me or post on HN.
  - buro9 15y ago
    I also agree. I started a proposal for Wiley publishing a long time ago on caching and networking for high performance web sites. Then I became employed by a company that forbade me from sharing (grrr) so I stopped. I felt when writing the proposal though, that the solution isn't just client, just network, just server, or just anything... it's the combination of all parts and knowing how a change in one place impacts another part and how to balance them all to achieve the goal. Even then, it's a stepped thing. A solution for a small but fast and resilient web site looks very different to a solution for a very large, distributed and fast web site. A book should really include the basics building blocks in all major areas, and then offer suggestions on how to apply those building blocks at different levels of scale.
    - Mikushi 15y ago
      Indeed. I am trying to involve a former colleague of mine who is a specialist in front-end development and optimization to complete what knowledge i have on this front. I have been actually thinking about asking HN or other about what they would like to see in this kind of book (keep in mind this is very PHP oriented as it's my specialty), instead of trying to come up with one myself and missing some important point. I am also thinking into recycling presentation i was hosting at internal IT gathering, on very wide subject range (internal working of PHP and how to take advantage of that knowledge, Varnish usage, HTML5 video, How to profile a PHP application with XHProf and what to look for), and make video presentation out of them. Or may be just bundling this in the book, not really sure yet.
jmilloy 15y ago
How is there an entire class devoted to "Rule 5: Put stylesheets at the top"? Yes, I clicked through the slides; it took 2 minutes.
- johnx123-up 15y ago
  Yep, I'm now skeptical about Stanford quality...
etherealG 15y ago
Anyone know where this falls in the standard computer science course?
- tikhonj 15y ago
  I'm guessing that this is more of a seminar-type class rather than a standard core CS course. It's very specialized and narrow and doesn't really fit into a standard CS education. I suspect the professor just really wanted to teach it, as opposed to the department deeming it a necessary part of the curriculum. Also, at my school classes starting with 19- are usually either seminars, student-taught or in "beta"; the numbering at Stanford might be similar. As an aside, some of these classes can be really awesome. Some nice examples at my school have included a popular cloud class and--particularly epic--a Starcraft-AI class. These are definitely not part of our core curriculum, but they are nice complements to staples like introduction to AI (the prerequisite for the Starcraft course).
  - etherealG 15y ago
    thanks for the clarification, that starcraft AI course sounds great! which school do you go to?
    - tikhonj 15y ago
      I go to Berkeley. The Starcraft AI class was structured around writing a bot to compete in a tournament which it ended up winning. I think there was a mix of graduate and undergraduate students. Unfortunately, I'm not sure they're doing it again.
dlauer 15y ago
I don't get it, where's the class on multi-tier server architecture? Load-balancing techniques? Server scaling? Database / backend performance best practices? I appreciate the focus on the front-end, and YSlow best practices, but I'm surprised that this kind of class is completely neglecting the backend - which is just as, if not more important, for scaling a major production website.
- chubot 15y ago
  Actually the entire premise of Steve's book is that he was in charge of performance/latency at Yahoo, and started out looking for issues in the backend. But when he measured end-user latency, he realized that all the issues were in the front end. It doesn't make sense to start tweaking 100,000 lines of C++ code to shave 1 ms from the backend response time when you're taking 200-300ms to load JavaScript and such. I think this is probably a common performance characteristic at large companies; perhaps not small ones. At small ones you probably have a database bottleneck. But at companies Yahoo/Google the backend engineers have generally optimized things quite a lot (partly because that kind of work is rewarded), and there's a lot of flab in the front end.

112 points by AbyCodes 15y ago

riffraff 15y ago

did anyone else just assume by default this was cs193h-class.org ?

blcArmadillo 15y ago
haha. yes.
- worldimperator 15y ago
  I guess after the first three weeks of ai-class.com, judging from the several stability adventures, the makers of the site could offer the knowledge of that course as well ;-)

cleverjake 15y ago

Steve Sounders is a wealth of information. His http archive (http://httparchive.org/ http://httparchive.org/) is already showing fundamental changes in design, as well as regularly giving good tips on high-scalability sites. I am jealous of those who get to take the full course. =]

wunderfool 15y ago

losing some respect for stanford's vaunted CS program here...this is devry territory. i would expect material for a stanford degree to be a bit more rigorous than regurgitating the recommendations of yslow.

prophetjohn 15y ago

I know it's not technically computer science and it's more vocational, but I wish my school offered classes like these. Something like this would be more beneficial to me than, say, Theory of Automata, I think.

lightcatcher 15y ago
really? I'm actually glad my school doesn't offer vocational classes like this. Optimizing a website on the client side is the sort of thing you read a blog post for and then go do. I don't enjoy it at all when I actually have to do it, I certainly wouldn't want to take a course where I have to do this.
- beagledude 15y ago
  that may be the newbest thing I've read on hacker news. It only shows you haven't worked in high performance environments.
  - lightcatcher 15y ago
    Do you believe that there aren't more useful things to learn in school? Stuff like this is fairly trivial to teach to yourself. I believe the time would be better served by taking a class in something like game theory or any one of the other mind-expanding courses offered (especially at a school like Stanford). As for the actual content of the class, all you really need to tell someone is that various things can make pages load slowly, and then link them to some guide such as this: http://developer.yahoo.com/performance/rules.html http://developer.yahoo.com/performance/rules.html . Also, I still have this opinion after working last summer in what I would consider a high performance environment at a YC company.
    - beagledude 15y ago
      I believe that CS courses should include more real world practices and situations. After interviewing hundreds of cs grads over the past decade it only re-enforces that belief.
    - simonw 15y ago
      I actually think high performance web architecture could make for an excellent mind-broadening course. Consider the things you need a deep understanding of: - DNS, how it works, DNS level load balancing, DNS client implementations in operating systems - Internet routing architecture and CDNs - Browser internals: how modern browsers access the network, how their parsers work, how they parallelise requests - In-depth understanding of both JavaScript and the DOM - How CSS style engines work - Different image file formats - How compression works - How to introspect and monitor web page performance - How to reverse-engineer complex web sites

mmaunder 15y ago

This isn't bad, but it's client focused. Perhaps there should be a second module that focuses on specific web server architectures (apache vs nginx and lighttpd), reverse proxying performance, performance of web servers talking to different data stores and the performance of those stores. Also server-side caching architectures as well as server hardware and what works best for web servers, caching servers and data stores.

Mikushi 15y ago
Agreed on that. I am actually currently working on a book (PHP oriented) on high performance / traffic website architecture. Worked 4 years at a Top50 Alexa and been eager to share my knowledge (was forbidden to talk publicly about most of it while i was working there).
- mmaunder 15y ago
  Sounds great, I'd love to learn more so email me or post on HN.
- buro9 15y ago
  I also agree. I started a proposal for Wiley publishing a long time ago on caching and networking for high performance web sites. Then I became employed by a company that forbade me from sharing (grrr) so I stopped. I felt when writing the proposal though, that the solution isn't just client, just network, just server, or just anything... it's the combination of all parts and knowing how a change in one place impacts another part and how to balance them all to achieve the goal. Even then, it's a stepped thing. A solution for a small but fast and resilient web site looks very different to a solution for a very large, distributed and fast web site. A book should really include the basics building blocks in all major areas, and then offer suggestions on how to apply those building blocks at different levels of scale.
  - Mikushi 15y ago
    Indeed. I am trying to involve a former colleague of mine who is a specialist in front-end development and optimization to complete what knowledge i have on this front. I have been actually thinking about asking HN or other about what they would like to see in this kind of book (keep in mind this is very PHP oriented as it's my specialty), instead of trying to come up with one myself and missing some important point. I am also thinking into recycling presentation i was hosting at internal IT gathering, on very wide subject range (internal working of PHP and how to take advantage of that knowledge, Varnish usage, HTML5 video, How to profile a PHP application with XHProf and what to look for), and make video presentation out of them. Or may be just bundling this in the book, not really sure yet.

jmilloy 15y ago

How is there an entire class devoted to "Rule 5: Put stylesheets at the top"? Yes, I clicked through the slides; it took 2 minutes.

johnx123-up 15y ago
Yep, I'm now skeptical about Stanford quality...

etherealG 15y ago

Anyone know where this falls in the standard computer science course?

tikhonj 15y ago
I'm guessing that this is more of a seminar-type class rather than a standard core CS course. It's very specialized and narrow and doesn't really fit into a standard CS education. I suspect the professor just really wanted to teach it, as opposed to the department deeming it a necessary part of the curriculum. Also, at my school classes starting with 19- are usually either seminars, student-taught or in "beta"; the numbering at Stanford might be similar. As an aside, some of these classes can be really awesome. Some nice examples at my school have included a popular cloud class and--particularly epic--a Starcraft-AI class. These are definitely not part of our core curriculum, but they are nice complements to staples like introduction to AI (the prerequisite for the Starcraft course).
- etherealG 15y ago
  thanks for the clarification, that starcraft AI course sounds great! which school do you go to?
  - tikhonj 15y ago
    I go to Berkeley. The Starcraft AI class was structured around writing a bot to compete in a tournament which it ended up winning. I think there was a mix of graduate and undergraduate students. Unfortunately, I'm not sure they're doing it again.

dlauer 15y ago

I don't get it, where's the class on multi-tier server architecture? Load-balancing techniques? Server scaling? Database / backend performance best practices? I appreciate the focus on the front-end, and YSlow best practices, but I'm surprised that this kind of class is completely neglecting the backend - which is just as, if not more important, for scaling a major production website.

chubot 15y ago
Actually the entire premise of Steve's book is that he was in charge of performance/latency at Yahoo, and started out looking for issues in the backend. But when he measured end-user latency, he realized that all the issues were in the front end. It doesn't make sense to start tweaking 100,000 lines of C++ code to shave 1 ms from the backend response time when you're taking 200-300ms to load JavaScript and such. I think this is probably a common performance characteristic at large companies; perhaps not small ones. At small ones you probably have a database bottleneck. But at companies Yahoo/Google the backend engineers have generally optimized things quite a lot (partly because that kind of work is rewarded), and there's a lot of flab in the front end.