The network situation has gotten pretty good compared to old 2G-ish baselines, even in the rural US, but oversubscription remains a major challenge even when you have "4G". I predict a 3Mbps connection will be a reasonable "slow" target over the next year or so.
-
-
Replying to @slightlylate @lmeyerov and
If you set a 5 second window to interactive, and ignore all connection setup and client processing costs, you can get something like 1.5MiB of data to a client in that window. But the clients and site designs make the target much harder in practice.
1 reply 0 retweets 1 like -
Replying to @slightlylate @lmeyerov and
Most sites I trace that have H/2 enabled *still* have (at least) two critical-path HTTP connections to set up. Assuming 2 connections and a P75-speed device, we cut that 1.5MiB to something like 500K. Each additional connection you need to set up subtracts an additional 50K.
1 reply 0 retweets 1 like -
Replying to @slightlylate @lmeyerov and
That P75 device is likely to have a Snapdragon 625; something like the Redmi S2: https://www.gsmarena.com/xiaomi_redmi_s2_(redmi_y2)-9185.php …
1 reply 0 retweets 1 like -
Replying to @slightlylate @lmeyerov and
P90-concerned sites -- folks that need to reach all users; think budget-market e-commerce, govt sites, etc. -- are on a much tougher budget this time next year. At the margin, networks aren't going to have even 3Mbps; so let's sub in 1.6Mbps as that's "fast 3G" today.
2 replies 0 retweets 1 like -
Replying to @slightlylate @lmeyerov and
The device is likely to be from a few years earlier too; say the Galaxy J2: https://www.gsmarena.com/samsung_galaxy_j2-7357.php … 4 28nm (slow) cores & very little RAM. Tough target today.
2 replies 0 retweets 2 likes -
It is frustrating that the US infra is behind most of the industrial world on internet.. and exactly why size opt for *beyond framework* matters, but not the stdlib. So afaict, it comes down to CPU time for a semi-decent ARM chip to lazy load 50KB... how much is it? 200ms?
1 reply 0 retweets 0 likes -
Cheap devices don't have "semi decent" ARM chips, they have the dregs of 2014's race to 64bit on a shitty 28nm process. They can't frequency scale, they have no cache, and their main-memory latency is *horrific*.
1 reply 0 retweets 0 likes -
Replying to @slightlylate @lmeyerov and
Bad process node & poor industrial design means they get throttled to prevent thermal runaway. Bad market dynamic means they get spun up hot only to stall on main memory. The P90 device is a long corridor with horrors behind every door.
1 reply 0 retweets 0 likes -
It'd be a super useful experiment to do a warm-cache load for an empty react page - cheap phone, industrial world phone, laptop. 28nm is 10X+ faster than from when I started desktop js app coding :)
1 reply 0 retweets 0 likes
Not sure what warm-cache load buys you there? If you want to do this, however, http://webpagest.org/easy can help.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
& Web Standards TL; Blink API OWNER
Named PWAs w/
DMs open. Tweets my own; press@google.com for official comms.