Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
LWN - How Google uses Linux (lwn.net)
36 points by philjr on Nov 11, 2009 | hide | past | favorite | 11 comments



It's a pity google keeps so much of the kernel secret sauce to itself, it would be nice if they gave back at least as much as they took from the kernel.


That's impossible, we don't change enough of it to give a similar amount back. We could give back more (and we're trying to get there), but as you read in the article much of what we do is back porting.


There might be plenty of other people interested in that back ported stuff. Why make the decision for them ?

By opening it up categorically you are bound to make a bunch of people very happy, and you might even 'inspire' the kernel maintainers to do the 'right thing'.


First of all, quite surprised Google is using perforce (woah!), but as the comments suggest, people internally are using a git-p4 bridge.

Secondly the hooks to let them tie I/O requests back to specific applications sounds funky, is there something in the wild that does this today?


I'm not. There are few source control systems capable of supporting development at google's scale, Perforce is one of them. It may be unhip (compared to modern DVCS systems), it may be a bit clunky (arguably it's less clunky than git), but it is definitely capable.


The impression I was left with was that their p4 stuff was for kernel source only (i.e. for the 30 or so developers that are using it), so I'm not sure if I see the big deal in using perforce in that scenario.


Their use is more widespread than that. They apparently have 3000 users and 100Gb of data on one primary server [1]

Although they have a source license so lord knows what they've done to the system to scale to their levels.

[1] http://www.versioncontrolblog.com/2006/12/03/perforce-as-the...


Well, consider that they've probably been using it for the better part of a decade.


What's wrong with Perforce? Sure it's "proprietary" but it's pretty solid and scales to huge organizations and codebases. I like having "one repository" but that's subjective I suppose.

I'm pretty sure DTrace can do that.


They can save on testing if they will post their patches to mainstream branch.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: