4 ms·

"We now insist that Web pages load in a quarter of a second, when we had no problem with two seconds in 2009 and four seconds in 2006." Not true at all. The "s

"We now insist that Web pages load in a quarter of a second, when we had no problem with two seconds in 2009 and four seconds in 2006." Not true at all. The "scientific" benchmark of "instant" page loads is 250ms. This is the goal. Pages should load instantly because they can load instantly, poor connection or large content notwithstanding. Four second page loads were absolutely not "okay" in 2006. They weren't even the norm. Always aim for those 250ms, everywhere. For example, if your programming workflow takes more than a second to do a build iteration cycle, it's bad. Slow tools are bad. The status quo may be bad, but that's no excuse not to aim higher.

nostrademons 8y ago
The parent comment is downvoted (pretty heavily, since an upvote still didn't un-grey it) as I reply, but it's absolutely correct. I was doing latency optimization for Google in 2009. There was a strong negative correlation between latency and click-through rate - enough that the additional latency of adding JQuery (~140ms IIRC) would've cost us about $150M, and just moving the ads above the results in document order was estimated at $50M (the latter figure turned out to be wrong when I ran the experiments in 2010 - broadband penetration had made the latency difference in document order negligible - but that was the figure Marissa Mayer got in 2002 when most users were on dialup, extrapolated out to Google's 2010 revenues). 2 seconds was absolutely too long; our target was 250ms for the whole search to come back, including all backend calls. If anything webpages have gotten slower today, with the rise of SPAs with lots of Javascript. Some of that is compensated for by faster Internet (though not a lot; in most of the U.S, speeds haven't budged much since 2009), but a lot is just that people have given up on the sort of marginal use-cases like amusing themselves that they used the web for in the 00s, and now tend to use it only for important things where they're willing to put up with slowness.
- andrewflnr 8y ago
  It's technically correct but irrelevant to the point the article is trying to make, which is more about tolerance of users for sub-par performance than about where the goal should be set. Granted, you don't have to be relevant. There's plenty of room for making side points as long as you don't try to present them as a takedown of the OP.
GordonS 8y ago
Yep, early web user here - I always had an issue with slow load times. After the WWW became a thing, I was on a 14k modem, and load times drove me up the wall. In 2006, load times still drove me up the wall - pages don't have to load in nanoseconds, but I want them to at least have the perception of loading instantaneously.
mindslight 8y ago
> Always aim for those 250ms, everywhere Erm no. that number varies depending on the context. For instance, the rule of thumb for UI lag on embedded devices is around 10ms to feel "instant". Which is why the current crop of surveillance devices feel like cheap junk, because their developers shrugged off that rule in favor of using latency-inducing technology stacks meant for servers. But not everything can be immediate either. For example, I'll take issue with build time. Sure if you're working with something that can get it down to near instantaneous that's great. But there's also a virtue in needing to plan the change you're making and what you expect it to do.
- maccard 8y ago
  > For example, I'll take issue with build time. Sure if you're working with something that can get it down to near instantaneous that's great. But there's also a virtue in needing to plan the change you're making and what you expect it to do. And I'll take issue with this too - my current work project takes about 4 minutes to _link_. That's the absolute quickest I can get. There is definitely a ride ground, and in my experience that middle ground is somewhere around the point of "do I check email/slack/hn when I hit build"
hodgesrm 8y ago
This line caught my attention too. It's hard to understand how anyone could argue that slower web pages are better. I typically have 50 to 100 open browser tabs at any given time and am constantly referencing material in them. Anything slower than a small fraction of a second is a material productivity hit.
- Izkata 8y ago
  It's not an argument that slower is better, but that the same duration was more acceptable to more people in the past. I do think they got the years wrong though; by memory 4 seconds load time was acceptable around 2000, not 2006.

nostrademons 8y ago

The parent comment is downvoted (pretty heavily, since an upvote still didn't un-grey it) as I reply, but it's absolutely correct. I was doing latency optimization for Google in 2009. There was a strong negative correlation between latency and click-through rate - enough that the additional latency of adding JQuery (~140ms IIRC) would've cost us about $150M, and just moving the ads above the results in document order was estimated at $50M (the latter figure turned out to be wrong when I ran the experiments in 2010 - broadband penetration had made the latency difference in document order negligible - but that was the figure Marissa Mayer got in 2002 when most users were on dialup, extrapolated out to Google's 2010 revenues). 2 seconds was absolutely too long; our target was 250ms for the whole search to come back, including all backend calls. If anything webpages have gotten slower today, with the rise of SPAs with lots of Javascript. Some of that is compensated for by faster Internet (though not a lot; in most of the U.S, speeds haven't budged much since 2009), but a lot is just that people have given up on the sort of marginal use-cases like amusing themselves that they used the web for in the 00s, and now tend to use it only for important things where they're willing to put up with slowness.

andrewflnr 8y ago
It's technically correct but irrelevant to the point the article is trying to make, which is more about tolerance of users for sub-par performance than about where the goal should be set. Granted, you don't have to be relevant. There's plenty of room for making side points as long as you don't try to present them as a takedown of the OP.

GordonS 8y ago

Yep, early web user here - I always had an issue with slow load times. After the WWW became a thing, I was on a 14k modem, and load times drove me up the wall. In 2006, load times still drove me up the wall - pages don't have to load in nanoseconds, but I want them to at least have the perception of loading instantaneously.

mindslight 8y ago

> Always aim for those 250ms, everywhere Erm no. that number varies depending on the context. For instance, the rule of thumb for UI lag on embedded devices is around 10ms to feel "instant". Which is why the current crop of surveillance devices feel like cheap junk, because their developers shrugged off that rule in favor of using latency-inducing technology stacks meant for servers. But not everything can be immediate either. For example, I'll take issue with build time. Sure if you're working with something that can get it down to near instantaneous that's great. But there's also a virtue in needing to plan the change you're making and what you expect it to do.

maccard 8y ago
> For example, I'll take issue with build time. Sure if you're working with something that can get it down to near instantaneous that's great. But there's also a virtue in needing to plan the change you're making and what you expect it to do. And I'll take issue with this too - my current work project takes about 4 minutes to _link_. That's the absolute quickest I can get. There is definitely a ride ground, and in my experience that middle ground is somewhere around the point of "do I check email/slack/hn when I hit build"

hodgesrm 8y ago

This line caught my attention too. It's hard to understand how anyone could argue that slower web pages are better. I typically have 50 to 100 open browser tabs at any given time and am constantly referencing material in them. Anything slower than a small fraction of a second is a material productivity hit.

Izkata 8y ago
It's not an argument that slower is better, but that the same duration was more acceptable to more people in the past. I do think they got the years wrong though; by memory 4 seconds load time was acceptable around 2000, not 2006.