Google's Numbers Every Developer Should Know

surana.wordpress.com

9 points·thedob··7 comments

Please don't post a link to your own blog when all it does is link to another post. Just link to the original.

Anyone know where the original is?

That's my blog, but I didn't post it here. The original link is in the post: http://moderator.appspot.com/#15/e=c9&t=10c

13th slide: http://cs343-spr0607.stanford.edu/materials/jeff_dean/

IIRC, this was a Stanford class, where they had invited a bunch of people to talk about software. The main site seems to be offline (http://cs343-spr0607.stanford.edu)

He may have meant this when he created the slide, but precise numbers here seem less important than relative magnitude as long as one is reasonably close.

One could be a few ns off on how long a L2 cache lookup takes, and endure in all but the most time-critical situations. Far bigger problems occur when one failed to account for drive seeks being around eight orders of magnitude slower than L2 cache lookups.

It doesn't strike me as a list of numbers to memorize and be able to recite, but more the sake of cultivating an intuition of where to start looking for places to optimize communication, and the relative order of things one should focus on.

"# Read 1 MB sequentially from network 10,000,000 ns # Read 1 MB sequentially from disk 30,000,000 ns"

That seems weird - wouldn't the 1MB be read from a disk at the other end of the network? So how can disk be slower than network (unless the network is all RAM-Disk...).

It doesn't have to be read from disk at the other end. It could be data that's already cached in memory, you just have to talk to the cache server over the network.

Reply on news.ycombinator.com