Behind the scenes, Tumblr is scaling at an enormous rate
As many know, Tumblr has been growing at a massive rate over the past year. This is a good thing of course. But it can also be bad because it leads to system strain and outages when you don’t have the scalability to account for the huge spike in usage. This is certainly something Tumblr has suffered through, with outages plaguing the site, it seems that Tumblr just hasn’t been able to keep up.
However, thanks to the eagle eye of Shortform Blog, some amazing stats were shared that come from an interview with Todd Huff from High Scalability and Tumblr systems engineer Blake Matheny.
Some of the technical content from the interview is way over my head, but there are some pearls in there that really give you a better idea of what is really happening behind the scenes at Tumblr to keep up with their massive growth and scale the site to an extent beyond imaginable so that issues like the Tumblr catastrophe are just a bad dream from a long time ago. Some key takeaways from the interview about Tumblr:
- 50+ million posts a day
- 500 million page views a day
- 15B+ page views month
- ~20 engineers
- Peak rate of ~40k requests per second
- 1+ TB/day into Hadoop cluster
- Many TB/day into MySQL/HBase/Redis/Memcache
- Growing at 30% a month
- ~1000 hardware nodes in production Billions of page visits per month per engineer
- Posts are about 50GB a day. Follower list updates are about 2.7TB a day.
- Dashboard runs at a million writes a second, 50K reads a second, and it is growing.
- 500 web servers
- 200 database servers
Another nugget in the interview that I noticed is that sometime in the future, Tumblr is going to expand on their inbox, adopting methodology similar to Facebook Messages. Currently Tumblr already has the Ask box, to send and ask questions of other users, and recently created Fan Mail, to send fan messages directly to other users. But in it’s current form, there is no user-to-user messaging system and if Tumblr moves to an inbox system, users could see some sort of peer messaging system in the future which would be really cool.
Some additional information from the interview is that Tumblr is re-writing the dashboard to scale further, because unlike most social websites, Tumblr is saying that “the graph for Tumblr users has hundreds of followers.” In comparison with Facebook who is at 850+ million users, the average user has 190 friends (or followers in this context). Tumblr’s social graph if accurate, makes Facebook’s social graph pale in comparison. Imagine if Tumblr was at Facebook’s scale, the average number of followers would be ten fold that amount.