API tools faq
paste
Advertisement
Guest User

SBUG UG GetMesh2 snippets

a guest
Sep 5th, 2013
75
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 6.05 KB | None | 0 0
raw download clone embed print report
  1. [22:08:52] Maestro Linden: all three RCs got monty's new project, which adds support for smarter asset fetching (which future viewers will use)
  2. [22:09:23] Maestro Linden: Monty had some details here: https://wiki.secondlife.com/wiki/Release_Notes/Second_Life_RC_BlueSteel/13#13.09.04.280593
  3. [22:10:23] Maestro Linden: GetMesh2 is like GetMesh, but pipelined
  4. [22:10:34] Maestro Linden: and it will use fewer connections
  5. [22:11:09] Maestro Linden: the throttling actions with 503 errors will mostly benefit high bandwidth, high-latency connections
  6. [22:12:26] Monty Linden: new RCs have a GetMesh2 cap that's mostly equivalent to GetMesh
  7. [22:12:36] Monty Linden: This is used to retrieve meshes.
  8. [22:12:49] Monty Linden: For reasons lost in time, viewers have been using far, far
  9. [22:12:58] Monty Linden: too many connections to do mesh downloads.
  10. [22:13:10] Monty Linden: 32 by default, extended over 500 by certain users
  11. [22:13:22] Monty Linden: completely unconstrained by certain debug/developer functions
  12. [22:13:40] Monty Linden: these have, umm, negatively impacted the experience for everyone.
  13. [22:13:54] Monty Linden: GetMesh2 is part of fixing this.
  14. [22:14:47] Monty Linden: On the server side, GetMesh2 now is serviced with
  15. [22:14:58] Monty Linden: keepalive connections and the viewer I'm currently using
  16. [22:15:00] Monty Linden: makes use of it.
  17. [22:15:35] Monty Linden: New texture console is going to have a line of status
  18. [22:15:43] Monty Linden: about mesh activities so you can see what it is doing.
  19. [22:16:06] Maestro Linden: Yuzuru, for GetMesh it's configurable viewer-side via 'MeshMaxConcurrentRequests' debug setting
  20. [22:16:13] Monty Linden: And there will be debug settings (Mesh2MaxConcurrentRequests or something like that)
  21. [22:16:36] Monty Linden: GetMesh2 is the cap/URL
  22. [22:16:56] Monty Linden: And there are other tools to monitor what viewers have been
  23. [22:16:57] Monty Linden: doing.
  24. [22:17:17] Monty Linden: You can kick up a 'netstat' command and filter for port 12046 activity to see what is happening
  25. [22:17:32] Monty Linden: on a viewer machine. It can be quite shocking when you
  26. [22:17:37] Monty Linden: TP to a mesh rich region.
  27. [22:17:50] Monty Linden: Or if you use certain features in the develop menu.
  28. [22:18:06] Monty Linden: One other interesting change is that some 503 responses
  29. [22:18:24] Monty Linden: from our web services may include a 'Retry-After' header with a delta value (simple seconds).
  30. [22:18:47] Monty Linden: Indicating a time at which it will be likely, but not guaranteed, for
  31. [22:18:51] Monty Linden: a new request to succeed.
  32. [22:19:11] Monty Linden: This applies to any request via cap subject to throttling including
  33. [22:19:13] Monty Linden: http-in.
  34. [22:19:27] Monty Linden: Though that's typically not the throttle of interest there.
  35. [22:20:10] Monty Linden: Rex: old and new viewers both did the downloads in a
  36. [22:20:25] Monty Linden: separate thread but there is a lot of mutex locking in the code.
  37. [22:20:33] Monty Linden: New viewer will have some interesting comments for those
  38. [22:20:37] Monty Linden: who want to read.
  39. [22:21:00] Monty Linden: I believe there are less stalls with the new code which cleaned up much but we'll see.
  40. [22:21:30] Monty Linden: The extreme cases can't even get in to services, they'll simply reset when they try to connect.
  41. [22:21:51] Monty Linden: Other connections will result in 503s and a connection closed by the server.
  42. [22:22:15] Monty Linden: Problem is that the original viewer coded a retry-as-fast-as-possible
  43. [22:22:22] Monty Linden: policy which made things worse.
  44. [22:22:30] Monty Linden: There is a pathological mesh case:
  45. [22:22:41] Monty Linden: retries happen so fast that *all* are 503'd and you never
  46. [22:22:46] Monty Linden: see any mesh progress.
  47. [22:22:55] Monty Linden: I'm curious if anyone here has experienced this.
  48. [22:23:01] Monty Linden: I can reproduce it fairly easily.
  49. [22:26:00] Monty Linden: the viewer is looking good and I'm trying to get it out
  50. [22:26:02] Maestro Linden: We should be clear that the viewer which respects Retry-After or uses the new GetMesh2 capability is not yet available. I believe the plan is to have a project viewer at some point?
  51. [22:26:03] Monty Linden: as quickly as possible.
  52. [22:26:16] Monty Linden: Also trying to get a better technical description out describing
  53. [22:26:23] Monty Linden: what I've done and what I'd like to do.
  54. [22:26:46] Monty Linden: All of this is preparation for HTTP pipelining which should be a very nice
  55. [22:27:02] Monty Linden: change, especially for high-ping-time locations (yuzuru).
  56. [22:27:35] Monty Linden: what maestro said, project viewer or maybe straight to RC
  57. [22:27:37] Monty Linden: not certain yet.
  58. [22:28:36] Monty Linden: that's the main thing
  59. [22:29:03] Monty Linden: are transmitted with 'chunked' encoding
  60. [22:29:13] Monty Linden: the value of this is that it provides redundant length information
  61. [22:29:22] Monty Linden: for reseponse data which allows libcurl and the viewer code
  62. [22:29:26] Maestro Linden: Ardy: Hm, 'motions' sounds like an avatar animation issue
  63. [22:29:31] Monty Linden: to validate that it really received a full response.
  64. [22:29:43] Monty Linden: If someone has a bad ISP or a bad network driver, they
  65. [22:29:51] Monty Linden: may see error messages indicating short data.
  66. [22:30:23] Monty Linden: (partial file transfer, or range header invalid)
  67. [22:30:37] Monty Linden: we'd like to hear about these but they likely indicate a
  68. [22:30:44] Monty Linden: weakness in the networking path.
  69. [22:32:13] Monty Linden: Rex, if actual data received on closed connection is shorter than that declared in content-length, you have a short response.
  70. [22:32:27] Monty Linden: Ditto if a connection closes in the middle of a chunk without completing the chunk.
  71. [22:32:37] Monty Linden: In fact, we have a jira where this happens.
  72. [22:33:23] Monty Linden: that's basically my story
  73. [22:34:00] Monty Linden: Anything inserting a RST into the stream will do it.
  74. [22:34:18] Monty Linden: Have seen bad HTTP caches that react badly to Range headers.
  75. [22:34:41] Monty Linden: It's a network. Things go wrong.
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!