Scaling Reddit from 1 Million to 1 Billion – Pitfalls and Lessons [video] • Hacker News Board

Re: Scaling Reddit from 1 Million to 1 Billion – Pitfalls and Lessons [video]

#111

Post by espeed » Sat, Aug 17, 2013, 5:13 AM UTC

billybob255 wrote: post #52

jlouis wrote: post #48

Earlier quoted context omitted.

It depends how much time is spent doing computation and how much time is spent doing lookups from disk. If the latter is clearly dominating, then a switch to Go will not help much.

I read a comment from one of the sysadmins (alienth?) saying their main bottlenecks were I/O.

With all of Reddit's past problems with AWS and EBS (https://news.ycombinator.com/item?id=2339214, https://news.ycombinator.com/item?id=2469838, ), I figured they would haved jumped to Google Compute Engine by now, which has much better IO all around (http://gigaom.com/2013/03/15/by-the-numbers-how-google-compu...).

Re: Scaling Reddit from 1 Million to 1 Billion – Pitfalls and Lessons [video]

#112

Post by jacques_chester » Sat, Aug 17, 2013, 6:46 AM UTC

diminish wrote: post #73

jlouis wrote: post #48

Earlier quoted context omitted.

It depends how much time is spent doing computation and how much time is spent doing lookups from disk. If the latter is clearly dominating, then a switch to Go will not help much.

thanks for the insight. i did not have much to do with Go. But hear a lot of positive benchmarks here on HN. Reddit is kind of like an example social web app, and I read a lot on its architectural changes and scaling efforts. So for social Web 2.0 apps, which are getting older now, I m curious how much Go would make a difference. Mainly for Google applications, apparently Go brings a lot of speed and performance on t…

It doesn't really matter what the genre of an application is. What matters is the runtime fundamentals. How much time is spent computing vs waiting for I/O? Whichever one is slower is the current bottleneck and is what you should fret about. Go becomes something to consider if the bottleneck is computation time. It's tangential otherwise.

Re: Scaling Reddit from 1 Million to 1 Billion – Pitfalls and Lessons [video]

#113

Post by grandpoobah » Sat, Aug 17, 2013, 8:59 AM UTC

jedberg wrote: post #54

grandpoobah wrote: post #39

Reddit still uses Pylons, and I imagine moving to Pyramid would be very painful. He's not clear on this in the video.

I don't think I made any indications one way or another, but yes, it would probably be difficult to move.

Great presentation man. Do you think Pylons will serve Reddit for years and years to come? Is there any need for you to switch to Pyramid or another framework? I made the choice to use Pylons for a recent project, and it just feels kind of odd using an old framework which is now in "maintenance only mode", but I truly did not like Pyramid... much less Django.

Re: Scaling Reddit from 1 Million to 1 Billion – Pitfalls and Lessons [video]

#114

Post by Houshalter » Sun, Aug 18, 2013, 4:35 AM UTC

amitparikh wrote: post #25

wjk wrote: post #16

Earlier quoted context omitted.

http://10minutemail.com/ is what i use, much simpler.

Another similar option is Mailinator: http://mailinator.com/

Some websites block emails from mailinator which is pretty annoying.