git notes main page | gitolite main page | license

IMPORTANT NOTE: although this page has a "gitolite.com" URL, this is not about gitolite. That's just an artifact of "sitaramc.github.com" being translated to "gitolite.com" and so ALL my git related stuff gets carried over. Gitolite documentation has another /gitolite in the URL, so you can tell. My apologies for this confusion.

One of the things git did was to try and optimise disk storage by adopting an aggressive delta compression technique. Most version control systems give you this, but they only do deltas for consecutive versions of the same file. Git not only tries to optimise across non-adjacent versions of a file (like if you removed a huge bunch of lines, and then added them back again), it also leverages similarities with other files.

The reason it does this is that it really tracks content, not just filenames.

Anyway, all this means that a packed git repo is much, much, smaller than a repo in any other VCS. (The Mozilla tree sizes, at one time, were:

One full checkout   :       350 MB
CVS repo            :       2.7 GB
SVN repo            :       8.2 GB
git repo            :       450 MB

Read that again: the entire history of Mozilla in barely a third more space than a normal checkout!

So what does all this have to do with performance? Well, although all this was done purely for disk space reasons, it turned out to have a surprising effect on performance.

It turned out (and in hindsight this was obvious) that, since the disk was the slowest component, keeping a small amount on the disk and making the CPU grunt a little was far, far, faster than not doing all this compression.

In fact, people found that a git checkout is faster than a “cp -a”, simply because of this tradeoff. Until you get used to it, this is a little mind-blowing.

Anyway the bottom line is, performance is not an issue.