[GHC] #9321: Support for waiting on multiple MVars

[GHC]

#9321: Support for waiting on multiple MVars
-------------------------------------+-------------------------------------
              Reporter:  schyler     |             Owner:  simonmar
                  Type:  feature     |            Status:  new
  request                            |         Milestone:
              Priority:  normal      |           Version:  7.8.3
             Component:  Runtime     |          Keywords:  mvar
  System                             |  Operating System:  Unknown/Multiple
            Resolution:              |   Type of failure:  None/Unknown
Differential Revisions:              |         Test Case:
          Architecture:              |          Blocking:
  Unknown/Multiple                   |
            Difficulty:  Unknown     |
            Blocked By:              |
       Related Tickets:              |
-------------------------------------+-------------------------------------

Comment (by schyler):

 I have been hacking together a benchmark for STM (attached) first, then
 later I will put together the equivalent thing for MVar. On my i7 haswell
 macbook pro I compile and run it like this:

 {{{
 $ ghc -O2 -threaded -fforce-recomp Bench.hs
 $ time ./Bench +RTS -N8
 }}}

 This emulates the exact use case I mentioned for the server above,
 actually. We have
 * A packets queue, bounded, written and read by only 1
 * An events chan, unbounded and written by everything (but individual
 forks)
 * A commands queue, bounded, written and read by 5% of the clients (but
 individual forks)

 In terms of the configurable parameters:
 * `threshold` is how many events, commands OR packets clients want before
 they just disconnect. I needed a fair simulation of throughput
 * `clients` is the number of clients to simulate connected to the server
 * There are adjustable delays on the client-local chans you can adjust,
 `packetDelay`, `eventDelay` and `commandDelay`. This is just to give a
 kind of estimate of the actual distribution of messages in a real server
 (I expect i.e. there to be 50 events per 1 actual client packet per 5
 client-specific commands)

 Interestingly the bench terminates in 4 seconds when `clients = 20`. I set
 `clients = 40` and couldn't get it to terminate for over 5 minutes (and my
 laptop was really hot) so I killed it.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9321#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler