summaryrefslogtreecommitdiffstats
path: root/src/smtpstone/hashed-deferred
diff options
context:
space:
mode:
Diffstat (limited to 'src/smtpstone/hashed-deferred')
-rw-r--r--src/smtpstone/hashed-deferred37
1 files changed, 37 insertions, 0 deletions
diff --git a/src/smtpstone/hashed-deferred b/src/smtpstone/hashed-deferred
new file mode 100644
index 0000000..93f0e12
--- /dev/null
+++ b/src/smtpstone/hashed-deferred
@@ -0,0 +1,37 @@
+Delivering 1000 deferred messages over the loopback transport,
+outbound concurrency 10. smtp-sink pipelining disabled. Machine is
+P230, BSD/OS 3.1, 64MB memory.
+
+hashing is 16 directories per level
+
+flat deferred queue
+
+ start: Sun Feb 21 16:42:37 EST 1999
+ done: Feb 21 16:44:35
+ time: 1:58 = 118 seconds
+
+ start: Sun Feb 21 16:48:01 EST 1999
+ done: Feb 21 16:49:51
+ time: 1:50 = 110 seconds
+
+hashed deferred queue, depth=1 (16 directories)
+
+ start: Sun Feb 21 17:29:36 EST 1999
+ done: Feb 21 17:31:32
+ time: 1:56 = 116 seconds
+
+ start: Sun Feb 21 17:33:36 EST 1999
+ done: Feb 21 17:35:24
+ time: 1:48 = 108 seconds
+
+ start: Sun Feb 21 17:37:08 EST 1999
+ done: Feb 21 17:39:02
+ time: 1:52 = 112 seconds
+
+Hashing does not slow down deliveries.
+
+However the problem is scanning an empty deferred queue. On an idle
+machine, it takes some 5 seconds to scan an empty depth=2 deferred
+queue unless the blocks happen to be cached. During those 5 seconds
+the queue manager will not pay attention to I/O from delivery
+agents, which is bad.