Decrease the number of send_pending_parcels threads

[brycelelbach]

[reported by blelbach] [Trac time Tue Feb 28 14:27:08 2012] |||GCC 4.6.2, 4b99001, Boost 1.49.0, debug

There are way too many send_pending_parcels threads. Running sheneos_test on two localities with default parameters creates about 27k pxthreads on the console, 11k of which are send_pending_parcels threads.

Steps to reproduce: Run sheneos_test on two localities with default parameters. Use logs to count number of threads like so:

cat my_hpx_log | grep thread::thread | wc -l
cat my_hpx_log | grep thread::thread | grep send_pending_parcels | wc- l

[brycelelbach]

[comment by heller] [Trac time Tue Feb 28 14:33:39 2012] In order to determine why, and if, send_pending_parcels are bad, we need network performance counters to fine tune the current connection caching, and parcel buffering.

[brycelelbach]

[comment by blelbach] [Trac time Tue Feb 28 15:19:36 2012] I don't see how this has anything to do with network performance counters, just decrease then number of send_pending_parcels threads.

In release, it's just as bad. 21k pxthreads, 9k of which are send_pending_parcels.

[brycelelbach]

[comment by blelbach] [Trac time Tue Feb 28 15:25:10 2012] BTW, in both debug and release I only send about 8k parcels total.

[brycelelbach]

[comment by heller] [Trac time Thu Mar 8 16:59:52 2012] This was fixed with e3f1307.