git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: Performance issue of 'git branch'

From
Linus Torvalds <torvalds@linux-foundation.org>
Date
Jul 23, 2009, 03:21 UTC
Message-ID
<alpine.LFD.2.01.0907222009340.21520@localhost.localdomain>
In-Reply-To
<alpine.LFD.2.01.0907221959330.21520@localhost.localdomain>
On Wed, 22 Jul 2009, Linus Torvalds wrote:
> 
> The GIT_DEBUG_LOOKUP debug output probably does match the number of 
> cold-cache IO's fairly well for something like this (at least to a first 
> approximation), so I really hope my patch will fix your problem.

Side note: the object lookup binary search we do is simple and reasonably efficient, but it is _not_ very cache-friendly (where "cache-friendly" also in this case means IO caches).

There are more cache-friendly ways of searching, although the really clever ones would require us to switch the format of the pack-file index around. Which would be a fairly big pain (in addition to making the lookup a lot more complex).

The _simpler_ cache-friendly alternative is likely to try the "guess location by assuming the SHA1's are evenly spread out" thing doesn't jump back-and-forth like a binary search does.

We tried it a few years ago, but didn't do cold-cache numbers. And repositories were smaller too.

With something like the kernel repo, with 1.2+ million objects, a binary search needs about 21 comparisons for each object we look up. The index has a first-level fan-out of 256, so that takes away 8 of them, but we're still talking about 13 comparisons. With bad locality except for the very last ones.

Assuming a 4kB page-size, and about 170 index entries per page (~7 binary search levels), that's 6 pages we have to page-fault in for each search. And we probably won't start seeing lots of cache reuse until we hit hundreds or thousands of objects searched for.

With soemthing like "three iterations of newton-raphson + linear search", we might end up with more index entries looked at, but we'd quite possibly get much better locality.

I suspect the old newton-raphson patches we had (Discussions and patches back in April 2007 on this list) could be resurrected pretty easily.

		Linus
Previous: Linus TorvaldsNext: Tony Finch
Message 14 of 73 in “Performance issue of 'git branch'”
  1. Carlos R. MafraJul 22, 2009
  2. Linus TorvaldsJul 23, 2009
  3. Linus TorvaldsJul 23, 2009
  4. Linus TorvaldsJul 23, 2009
  5. Carlos R. MafraJul 23, 2009
  6. Linus TorvaldsJul 23, 2009
  7. Jakub NarebskiJul 23, 2009
  8. Carlos R. MafraJul 23, 2009
  9. Linus TorvaldsJul 23, 2009
  10. Carlos R. MafraJul 23, 2009
  11. Linus TorvaldsJul 23, 2009
  12. Linus TorvaldsJul 23, 2009
  13. Linus TorvaldsJul 23, 2009
  14. Linus TorvaldsJul 23, 2009
  15. Tony FinchJul 23, 2009
  16. Linus TorvaldsJul 23, 2009
  17. Newton-Raphson, was Re: Performance issue of 'git branch'Tony Finch, Jul 23, 2009
  18. Johannes SchindelinJul 23, 2009
  19. Tony FinchJul 23, 2009
  20. Johannes SchindelinJul 24, 2009
  21. Carlos R. MafraJul 23, 2009
  22. Carlos R. MafraJul 23, 2009
  23. Carlos R. MafraJul 23, 2009
  24. Linus TorvaldsJul 23, 2009
  25. Linus TorvaldsJul 23, 2009
  26. Junio C HamanoJul 23, 2009
  27. Carlos R. MafraJul 23, 2009
  28. Junio C HamanoJul 23, 2009
  29. Linus TorvaldsJul 23, 2009
  30. Junio C HamanoJul 23, 2009
  31. Junio C HamanoJul 23, 2009
  32. Linus TorvaldsJul 23, 2009
  33. Carlos R. MafraJul 23, 2009
  34. Linus TorvaldsJul 23, 2009
  35. Carlos R. MafraJul 23, 2009
  36. Linus TorvaldsJul 23, 2009
  37. Linus TorvaldsJul 23, 2009
  38. Carlos R. MafraJul 23, 2009
  39. Linus TorvaldsJul 24, 2009
  40. Linus TorvaldsJul 24, 2009
  41. Linus TorvaldsJul 24, 2009
  42. Linus TorvaldsJul 24, 2009
  43. david@lang.hmJul 24, 2009
  44. Linus TorvaldsJul 24, 2009
  45. david@lang.hmJul 24, 2009
  46. Linus TorvaldsJul 25, 2009
  47. Daniel BarkalowJul 25, 2009
  48. Jeff KingAug 7, 2009
  49. Theodore TsoJul 24, 2009
  50. Shawn O. PearceJul 24, 2009
  51. Junio C HamanoJul 24, 2009
  52. Avi KivityJul 26, 2009
  53. Johannes SchindelinJul 26, 2009
  54. Carlos R. MafraJul 24, 2009
  55. Carlos R. MafraJul 25, 2009
  56. Linus TorvaldsJul 25, 2009
  57. Timo HirvonenJul 25, 2009
  58. Reece DunnJul 25, 2009
  59. Mike HommeyJul 25, 2009
  60. Linus TorvaldsJul 25, 2009
  61. Linus TorvaldsJul 25, 2009
  62. Johannes SchindelinJul 25, 2009
  63. Linus TorvaldsJul 26, 2009
  64. Theodore TsoJul 26, 2009
  65. Mike HommeyJul 26, 2009
  66. Johannes SchindelinJul 26, 2009
  67. demerphqJul 26, 2009
  68. demerphqJul 26, 2009
  69. Carlos R. MafraJul 25, 2009
  70. Anders KaseorgJul 23, 2009
  71. Carlos R. MafraJul 23, 2009
  72. SZEDER GáborJul 23, 2009
  73. Carlos R. MafraJul 23, 2009

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.