Author: toad
Date: 2008-04-08 14:27:58 +0000 (Tue, 08 Apr 2008)
New Revision: 19074
Modified:
trunk/freenet/src/freenet/node/LocationManager.java
Log:
Increase SWAP_RESET to 16000. Factor of 4 so we should see obvious results -
less location churn, if this matters.
Explain the alchemy.
Modified: trunk/freenet/src/freenet/node/LocationManager.java
===================================================================
--- trunk/freenet/src/freenet/node/LocationManager.java 2008-04-08 14:26:46 UTC
(rev 19073)
+++ trunk/freenet/src/freenet/node/LocationManager.java 2008-04-08 14:27:58 UTC
(rev 19074)
@@ -60,8 +60,14 @@
static final int TIMEOUT = 60*1000;
static final int SWAP_MAX_HTL = 10;
/** Number of swap evaluations, either incoming or outgoing, between
resetting our location.
- * There is a 2 in SWAP_RESET chance that a reset will occur on one or
other end of a swap request. */
- static final int SWAP_RESET = 4000;
+ * There is a 2 in SWAP_RESET chance that a reset will occur on one or
other end of a swap request.
+ *
+ * ALCHEMY: This depends on a number of factors, not least the size of the
network. It is hard to
+ * get a good value from simulations. But it can take time to recover
after a random reset, so we
+ * have increased it from 4000 to 16000 on 8 april 2008. At the time
location churn was significant,
+ * and some of it likely caused by this. OTOH if we get major keyspace
fragmentation, we must
+ * reduce it to 8000 or 4000. */
+ static final int SWAP_RESET = 16000;
// FIXME vary automatically
static final int SEND_SWAP_INTERVAL = 8000;
/** The average time between sending a swap request, and completion. */
