Re: TIMEOUT_?

Bob Purdon writes:

>Almost no timeouts before - maybe 1-2%. We're seeing around (guesstimate)
>80% timeouts now. Just about every ICP query the sibling cache receives
>is being reported as a timeout by the requesting cache :-(

Try this patch below if you like. It uses double the maximum RTT instead
of the average. Let me know how it affects service times.

Index: neighbors.c
===================================================================
RCS file: /surf1/CVS/squid/src/neighbors.c,v
retrieving revision 1.220
diff -w -u -r1.220 neighbors.c
--- neighbors.c 1998/06/03 20:34:44 1.220
+++ neighbors.c 1998/06/05 07:34:14
@@ -490,7 +490,12 @@
         } else if (neighborUp(p)) {
             /* its alive, expect a reply from it */
             (*exprep)++;
+#if AVERAGE
             (*exprtt) += (double) p->stats.rtt;
+#else /* MAX */
+ if ((double) p->stats.rtt > (*exprtt))
+ (*exprtt) = (double) p->stats.rtt;
+#endif
         } else {
             /* Neighbor is dead; ping it anyway, but don't expect a reply */
             /* log it once at the threshold */
@@ -544,8 +549,13 @@
     /*
      * Average out the expected RTT and then double it
      */
+#if AVERAGE
     if (*exprep > 0)
         (*exprtt) = 2.0 * (*exprtt) / (double) (*exprep);
+#else /* MAX */
+ if (*exprep > 0)
+ (*exprtt) *= 2.0;
+#endif
     else
         *exprtt = Config.neighborTimeout;
     return peers_pinged;

Duane W.
Received on Fri Jun 05 1998 - 00:36:31 MDT