[netperf-dev] netperf2 commit notice r49 - trunk/src
raj at netperf.org
raj at netperf.org
Tue Jan 31 12:48:57 PST 2006
Author: raj
Date: 2006-01-31 12:48:54 -0800 (Tue, 31 Jan 2006)
New Revision: 49
Modified:
trunk/src/nettest_bsd.c
Log:
make the new and improved first burst functionality work correctly when
confidence intervals are being used.
Modified: trunk/src/nettest_bsd.c
===================================================================
--- trunk/src/nettest_bsd.c 2006-01-31 07:43:54 UTC (rev 48)
+++ trunk/src/nettest_bsd.c 2006-01-31 20:48:54 UTC (rev 49)
@@ -4929,6 +4929,7 @@
struct tcp_rr_results_struct *tcp_rr_result;
#ifdef WANT_FIRST_BURST
+#define REQUEST_CWND_INITIAL 2
/* "in the beginning..." the WANT_FIRST_BURST stuff was like both
Unix and the state of New Jersey - both were simple an unspoiled.
then it was realized that some stacks are quite picky about
@@ -4939,12 +4940,17 @@
happen that frankly, we cannot guarantee with the specification
of TCP. ain't that grand?-) raj 2006-01-30 */
int requests_outstanding = 0;
- int request_cwnd = 2; /* we ass-u-me that having three requests
- outstanding at the beginning of the test
- is ok with TCP stacks of interest. the
- first two will come from our first_burst
- loop, and the third from our regularly
- scheduled send */
+ int request_cwnd = REQUEST_CWND_INITIAL; /* we ass-u-me that having
+ three requests
+ outstanding at the
+ beginning of the test
+ is ok with TCP stacks
+ of interest. the first
+ two will come from our
+ first_burst loop, and
+ the third from our
+ regularly scheduled
+ send */
#endif
#ifdef WANT_INTERVALS
@@ -5005,6 +5011,15 @@
timed_out = 0;
trans_remaining = 0;
+#ifdef WANT_FIRST_BURST
+ /* we have to remember to reset the number of transactions
+ outstanding and the "congestion window for each new
+ iteration. raj 2006-01-31 */
+ requests_outstanding = 0;
+ request_cwnd = REQUEST_CWND_INITIAL;
+#endif
+
+
/* set-up the data buffers with the requested alignment and offset. */
/* since this is a request/response test, default the send_width and */
/* recv_width to 1 and not two raj 7/94 */
More information about the netperf-dev
mailing list