No One Tweets Like the Japanese, and That Was a Problem for Twitter

Twitter's engineering blog posted about this phenomenon last year.[1] It's not as in-depth as I'd like, but it does contain some technical/architectural details about how they scaled to deal with such high traffic.

1. https://blog.twitter.com/2013/new-tweets-per-second-record-a...

Didn't read the article, but I did grep it for unicode and utf-8 and found no matches, so this might also be related.

Twitter lets you tweet 140 characters regardless of the bit width of those characters in. For Japanese, I think almost all characters take two bytes in utf-8. As such, given the same number of tweets, the bandwidth usage is approx 2X.

Twitter also seems much more useful in ideogram languages as 140 characters = 140 words = an article. In English, 140 characters = a short/medium sized sentence.

"built a new system—known as a software “framework,” in engineering speak"

Hearing the words "engineering speak" bothers me. Wired knows a significant portion of its readers are engineers, right?

Saying just software framework would have probably been fine. Humans are pretty good at inferring the meaning of words from context...

Japanese is often encoded in Shift_JIS, which is much better than UTF-8 (for japanese text). Most browsers default to that encoding on japanese OSes, so there probably isn't a real bandwidth issue, depending on the ratio of browser to dedicated client usage.

As for ideograms, your statement is not correct. Kanjis allow for better content/character ratios, certainly, but one kanji is very often not one word - the word foreigner, for example, uses 3 ideograms.

On top of that, Japanese is not written solely with kanjis (as opposed to Chinese, for example). It also uses katakanas and hiraganas, which stand for phonems. This is more often the case on social networks where a lot of western words are used - western words are almost systematically written with katakanas.

Kanas are still more "efficient" than alphabets, but to reuse my previous example, foreigner is written using 5 kanas instead of 3 kanjis.

I remember reading a feature in Wired 10+ years ago about how processors were made that went into some quantum details and basically never stopped to take a breath. It was refreshing to see a publication simply assume you were smart enough to read, and thus smart enough to follow along (or look up the hard words yourself). I guess times have changed a bit.