Some of these tips are really good (using a CDN, caching on basically every layer, using YSlow), but does the author really believe using FPGAs and ASICs for websites is really practical? I've yet to see one real-world report of people using FPGAs to speed up sites. Hell, I haven't even seen any academic proof of concepts focusing on the merits of FPGAs in the web environment.
That's not to say that FPGA's won't be useful (they could potentially be helpful for real-time ranking and the like), it just seems out of character to mention something like that on a website scalability site. Grain of salt, and all that.
I think it's the web nerds equivalent of mechanics talking about Formula 1 race cars, most people who are in the field will never even be closing to needing that level of technology, but when you do you really do.
If you are doing lots of ssl connections some form of accelerator, whether drop in cards or a standalone box, are well worth the cost.
It is overkill, but dedicated special purpose hardware like a hardware SSL Accelerator (http://en.wikipedia.org/wiki/SSL_acceleration) is certainly useful, and its not unfathomable that one of these would use an FPGA.
FPGAs are impractical. They are designed for integration and custom design rather than performance. Any data processing tasks are probably not worth it. Intel i7 series CPUs are probably a better investment as they have native text-processing instructions!
I just found a benchmark on Intel's site that shows the effect of enabling the text processing instructions in SSE4.2. They claim a 25% speedup for XML parsing and 37% for XML validating.
I think this quote should have been a headline as opposed to an easily missable list item
'80-90% of the end-user response time is spent on the
frontend, so it makes sense to concentrate efforts there
before heroically rewriting the backend.'
Additionally, and this is a point often made by the YSlow guys (I am full of fanboy worship for them since they have made me thousands): Optimizing the black end is black magic full of interventions that you're never quite sure will work until you try them (and sometimes not even then), but optimizing the front end is a science. Front end best practices can be objectively demonstrated to work and they are repeatable.
Example: if your site loads 3 CSS files, and you combine them into 1 CSS file, you will experience performance increases. Period. If you do not gzip your HTML/Javascript/CSS yet, and you are not making a pathological example site just to prove me wrong, you will experience (probably significant!) performance increases. If you increase the number of static assets you load in parallel (for example, by splitting static assets over 2 additional subdomains), you will see most browsers have significant, automatic performance increases for pages that load non-trivial amounts of static assets. etc, etc
Watch the YSlow guys and be enlightened. Everything they produce about the subject (and I don't just mean the plugin) is solid gold. They are easily the best "Follow these simple directions and you will make money" technical material I've ever seen.
Get the two Souders (creator of YSlow, now works on web performance at Google) books, which are amazing. I'm not affiliated with Amazon.com, O'Reilly or Yahoo!; just like Patrick says, these guys know their shit:
A bit off topic, but could someone recommend any good articles or case studies highlighting the importance of site design and the potential affect on sales of nuanced site changes (ie: I recall reading an article about how a small change to a checkout process increased sales x%, but can't find that article).
I am trying to put together some reading material for a non-technical person in my startup who is unconcerned that our web design company is botching things.
(Even if you don't have a specific link, if you remember reading something and can remember the company name or an googleable information, it would be much appreciated).
Most of my articles are on my blog rather than my business site -- I only put them there when I need to use Rails to spit something out for the article.
That's not to say that FPGA's won't be useful (they could potentially be helpful for real-time ranking and the like), it just seems out of character to mention something like that on a website scalability site. Grain of salt, and all that.