CPU usage higher than normal
-
I've noticed since moving to RC2, that the CPU usage on our PFSense box is higher than it normally is. I'm not sure if this is cause for alarm or not, but it has stopped responding 3 times in 3 weeks (each time was almost exactly a week apart).
I didn't do any debugging, as this is in a production environment, and getting it back up and running ASAP was of the utmost importance. Power cycle, and everything comes back up fine.
Little bit about the system - Dell Poweredge 350, P3 850, 512MB RAM, 40GB HDD, HDD install.
We currently use it as a filtered bridge for our webhosting company, and it is filtering a consistent 5-10mbit of traffic that is nearly all web/mail traffic.
Immediately after a reboot, we see CPU usage in the 15-25% range (as reported by the WebConfigurator). After one day, we are seeing spikes at 60 and 70%, with agerages around 30-35%. Memory usage is low, 8-9%, and our state table has been increased to 100k entries, with anywhere from 10-15k states, set to aggressive.
Is this something that we need to look in to as a possible hardware problem, or is this something that could have been introduced with RC2? Is this normal with the amount of traffic we are seeing? We've been using m0n0wall as our primary firewall previously, and we had excellent system uptime on the same hardware.
We also run an identical system to load balance SpamAssassin requests across a cluster of BSD computers on our network. That system is also configured with RC2, but only receives about .5 -1mbit of sustained traffic. The uptime on that system is at 16 days (since we installed RC2) and prior to that, RC1 ran from the time we installed it, from the time we updated to RC2.
Any thoughts or suggestions as to the issues we are seeing would be greatly appreciated.
Matt Breitbach
-
When the cpu usage climbs again in one week please run top on the firewall to see what is taking up the cpu ticks.
-
As long as I can get into it, I'll do that.
-
You don't have console access?
-
I didn't check actual console access, but the webgui has been completely frozen each time I've gone in and rebooted it. Again, when it went down, we were more concerned with getting it back up and running than debugging the problem.
When it goes down again, I'll probably just route around it (it's just a filtered bridge, so swapping e-net cables around isn't a big deal) and then debug.
-
I have exactly the same problem. Mhen the traffic exeeded 3 Mb/s the CPU is on 100 % which limits the speed. I logged through SSL and start TOP then start a download. CPU load for interrupst never drop bellow 25 % and after the start it goes up to 92 %. But I found something interesting when turn off Pooling the load drops to 10-11% which is good but after the start of bigger download goes to 100 % due to interrupts. I am using the latest snapshot. I have no problems with m0n0wall during such loads and i achieve even 10 Mb/s with no problems.
-
The latest update features some bpf fixes that may help with this.
http://www.pfsense.com/~sullrich/1.0-SNAPSHOT-09-26-06/
-
In case it can help, on my pfSense RC2 when I run top syslogd is taking about 98% of CPU.
-
I have exactly the same problem. Mhen the traffic exeeded 3 Mb/s the CPU is on 100 % which limits the speed. I logged through SSL and start TOP then start a download. CPU load for interrupst never drop bellow 25 % and after the start it goes up to 92 %. But I found something interesting when turn off Pooling the load drops to 10-11% which is good but after the start of bigger download goes to 100 % due to interrupts. I am using the latest snapshot. I have no problems with m0n0wall during such loads and i achieve even 10 Mb/s with no problems.
What's the processor in your box? This sounds like classic signs of maxing out the capacity of your CPU.
Having it work with m0n0wall doesn't mean anything. m0n0wall is 30-50% faster on low end CPU's, network throughput wise, than pfsense because of the difference between FreeBSD 4.11 and 6.1. m0n0wall will see the same drop off when it gets to 6.1.
The other problems mentioned in this thread seem to be legit issues (especially syslogd pegging it at 100%), but I don't think this one is.
-
The syslogd issue comes up from time to time. It is a errata item for 1.0. Unfortunately I have not been able to locate the problem. When it happens simply click the save button in the logging settings section.
-
My problem is overpowered CPU on 266 Mhz ;) must change the system if have problems with more traffic for now it is not issu\
Edit: the numbers from the previous post are time 10 bigger it is for 30 Mbit/s when pfsense is at 100 % not 3 with m0n0wall top speed with the same system is about 50-60 Mbit/s. 30 Mbit/s is more than enough for the current use :) 10x for reading.
-
My problem is overpowered CPU on 266 Mhz ;) must change the system if have problems with more traffic for now it is not issu\
Edit: the numbers from the previous post are time 10 bigger it is for 30 Mbit/s when pfsense is at 100 % not 3 with m0n0wall top speed with the same system is about 50-60 Mbit/s. 30 Mbit/s is more than enough for the current use :) 10x for reading.
A 266 MHz, if we're talking about a WRAP or Soekris 4801, will push about 30-32 Mbps with pfsense, and 40-45 Mbps with m0n0wall. If it's a generic PC with a 266 MHz proc, numbers will vary depending on what type of machine, what NIC's, etc.
Those numbers are for an optimal transfer - single TCP stream, optimal packet size, etc.