We have put it into the production scenario (2.0b6) last night.
Unfortunaly, we didn't benchmark the Alteon load balancer...
I hope to do some tests agains Pound in the production scenario but the
results will be similar with the ones I already sent to the list: with
two P3 500MHz, 512MB RAM, balancing to 2 servers, the throughput was
almost 70 requests per second, without syslog logs. With syslog logs
enabled, it came to thirty few requests per second.
Some interesting points though:
1. LB seems to be doing fine the balancing, with almost completely
distributed loading (see statistics bellow)
2. CPU Load is high with "LogLevel 2" active in Pound (see below) due to
syslog activity and IO wait; disabling logs with "LogLevel 0" takes the
load average down to ; Syslog is configured to log local2. to
"-/var/log/pound".
3. after "bad header" messages are removed, I hope to see less IO
activity since all requests dump a lot of trash to syslog
=== Partial "top" output with "LogLevel 2" in pound.cfg ===
top - 14:46:16 up 23:55, 3 users, load average: 2.09, 2.01, 1.84
Tasks: 47 total, 1 running, 46 sleeping, 0 stopped, 0 zombie
Cpu(s): 3.6% us, 1.3% sy, 0.0% ni, 24.4% id, 70.0% wa, 0.7% hi,
0.0% si
Mem: 515708k total, 506204k used, 9504k free, 48696k buffers
Swap: 522104k total, 0k used, 522104k free, 390804k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
20073 root 15 0 28392 6644 1364 S 3.3 1.3 36:01.12 pound2
179 root 15 0 0 0 0 D 0.3 0.0 1:31.98 kjournald
1813 root 15 0 1608 628 536 D 0.3 0.1 2:05.53 syslogd
=== Partial "top" output with "LogLevel 0" in pound.cfg ===
top - 14:52:02 up 1 day, 1 min, 3 users, load average: 0.08, 0.99,
1.46
Tasks: 47 total, 1 running, 46 sleeping, 0 stopped, 0 zombie
Cpu(s): 2.3% us, 0.7% sy, 0.0% ni, 96.7% id, 0.0% wa, 0.3% hi,
0.0% si
Mem: 515708k total, 505624k used, 10084k free, 48704k buffers
Swap: 522104k total, 0k used, 522104k free, 394232k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
28789 root 15 0 9876 2744 1364 S 2.3 0.5 0:04.73 pound2
1813 root 16 0 1608 628 536 S 0.0 0.1 2:05.92 syslogd
==== STATISTICS ====
Per day:
------------
[root(at)dino-lb1 log]# fgrep "OK (10.112.70.21" pound.log | cut -c 1-6 |
uniq -c
15687 Jan 27
[root(at)dino-lb1 log]# fgrep "OK (10.112.70.22" pound.log | cut -c 1-6 |
uniq -c
15309 Jan 27
[root(at)dino-lb1 log]#
Per hour:
--------------
[root(at)dino-lb1 log]# fgrep "OK (10.112.70.21" pound.log | cut -c 1-10
| uniq -c
3209 Jan 27 00:
2408 Jan 27 01:
1552 Jan 27 02:
973 Jan 27 03:
477 Jan 27 04:
457 Jan 27 05:
430 Jan 27 06:
1103 Jan 27 07:
1743 Jan 27 08:
2767 Jan 27 09:
276 Jan 27 10:
[root(at)dino-lb1 log]# fgrep "OK (10.112.70.22" pound.log | cut -c 1-10
| uniq -c
4103 Jan 27 00:
2169 Jan 27 01:
1620 Jan 27 02:
1244 Jan 27 03:
451 Jan 27 04:
380 Jan 27 05:
355 Jan 27 06:
749 Jan 27 07:
1409 Jan 27 08:
2538 Jan 27 09:
164 Jan 27 10:
[root(at)dino-lb1 log]#
Regards,
Sergio Freire
-----Original Message-----
From: Robert Segall [mailto:roseg(at)apsis.ch]
Sent: Friday, January 27, 2006 1:54 PM
To: pound(at)apsis.ch
Subject: Re: [Pound Mailing List] Problem with Pound 2.0b6
On Thu, 2006-01-26 at 16:15 +0000, Sergio Freire wrote:
[...]
[...]
[...]
[...]
[...]
[...]
etc...
OK - it will be fixed in the 2.0 final release.
Any news on your benchmarks?
[...]
|
