| commit | d4353339d72a0f918ad2372b8f5af99b973f56cb | |
| author | Shawn O. Pearce <spearce@spearce.org> | |
| Sat, 27 Dec 2008 17:32:51 +0000 (27 09:32 -0800) | ||
| committer | Robin Rosenberg <robin.rosenberg@dewire.com> | |
| Sun, 4 Jan 2009 00:05:33 +0000 (4 01:05 +0100) | ||
| tree | d38996ca098316668534693c6e3a00482d4708d5 | treesnapshot (tar.gz zip) |
| parent | 20f2172d133658905d318c36562172901bdd7854 | commitdiff |
Rewrite WindowCache to use a hash table
This shaved 10s off my linux-2.6 "jgit rev-list --objects" test.
We're also getting slightly better cache hit ratios, approaching
95% during commit access and 80% during tree access on the same
repository. Its a significant win.
The cache is also a bit better about memory usage. We now make
sure the byte[] or ByteBuffer handle is cleared out of the soft
reference when we push the reference into the queue manually. A
small change, but it makes a marked difference with the overall
JVM memory usage.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
This shaved 10s off my linux-2.6 "jgit rev-list --objects" test.
We're also getting slightly better cache hit ratios, approaching
95% during commit access and 80% during tree access on the same
repository. Its a significant win.
The cache is also a bit better about memory usage. We now make
sure the byte[] or ByteBuffer handle is cleared out of the soft
reference when we push the reference into the queue manually. A
small change, but it makes a marked difference with the overall
JVM memory usage.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>