[tor-bugs] #12890 [Tor]: Design and implement optimizations for socket write limits

[Tor Bug Tracker & Wiki]

#12890: Design and implement optimizations for socket write limits
-----------------------------+--------------------------------
     Reporter:  robgjansen   |      Owner:  robgjansen
         Type:  enhancement  |     Status:  new
     Priority:  normal       |  Milestone:  Tor: 0.2.6.x-final
    Component:  Tor          |    Version:
   Resolution:               |   Keywords:  tor-relay
Actual Points:               |  Parent ID:  #12541
       Points:               |
-----------------------------+--------------------------------

Comment (by robgjansen):

 Replying to [comment:11 nickm]:

 > (Rob, please let me know which of the above are necessary before you
 could start testing this.)

 Some changes and thoughts:

  1. **I have attached a patch** for a slight refactoring of the thread
 main loop to make it easier to run the thread in Shadow.
  1. I thought about it, and I'm pretty sure we don't actually want the
 socket size calculation anymore. I needed it in my prototype because I was
 not setting socket buf sizes explicitly, and so had to deal with the buf
 size changing underneath me from tcp autotuning. Since we are setting the
 buf size, taking `buf_size = MIN(sock_space, tcp_space);` would mean that
 the buf size could never increase - and I think we don't want that! So,
 all we would have to compute is `tcp_space` and we can save the other two
 syscalls :) **I have attached a patch for this as well.**
  1. It would be useful to have the global read/write limits feature there
 before starting to test. For now, this could be as simple as a
 configuration option that specifies the upstream link capacity, and then
 the logic necessary to enforce the limit.
  1. We probably also want to write a bit above the per-socket and/or
 global limits, so that we can minimize kernel 'starvation' (when the
 kernel could have sent something, but it had no data to send because our
 calculations were a bit off). Options that specify how much above the per-
 socket and global limits we want to write would be helpful. For example,
 `buf_size = tcp_space;` would become `buf_size = tcp_space *
 get_options()->KISTSockBufSizeFactor;` (e.g., 1.25) and there would be a
 similar config for global limits. I think upper/lower bounds are a good
 idea to enforce some max/min values.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/12890#comment:12>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online