Hacker News Reader

Comment by oliverkofoed | Hacker News Reader

Google is also one of a very few companies where this type of micro-micro-optimizing even begins to make sense.

No it doesn't: this makes sense for everyone. Scale would be involved if they were out to save a few bytes per request, to save on bandwidth. However, they are out to lower the per user load/rendering time, which is completely independent of the scale of the company delivering the page (assuming the delivery scales well enough that extra traffic does not lower page delivery time, but that is usually the case).

I think he's right. Google's home HTML is only 5kB, which is pretty low these days, and their images and CSS aren't much more. If your page size is less than 10kB, those couple of bytes start making a difference. Google have probably literally optmised everything else they can optimise. Image atlas to reduce it to one file, they have a CDN, Custom web server, custom OS kernel, etc., some of which helps far more than some closing tag.

Yes, all those things are determined by scale, because there is a clear cost involved, that requires a certain scale to pay for itself. However, in the case of closing tags the cost is negligible, so it's something that can be used by everyone. Google claims (however, I still can't find the link that shows) that loading/rendering their search result pages slower makes people perform less searches. That's why they do not allow you to have more than 10 results per page; not even via your personal settings. The same trade off will hold for every service whose profitability is directly dependent on the number of page views. So it's not scale, just the sort of business you are in that determines whether not closing tags makes sense.

Am I wrong or are people convinced the parent must be right, given he's got 50 upmods? I was kind of hesitant to criticise it with that many upmods, but really, am I that obtuse if I don't see his point?

pmjordan pointed out that my assertion that the per user loading/rendering time is completely independent of the scale is overstating the case, as he illustrated with a CDN and custom software. However, in my response to him, I think I made a good case that that argument doesn't hold for stripping closing tags?

Reply on news.ycombinator.com