Forum Discussion
NimbostratusPool round-robin load balancing not balanced
I have a simple pool created with 4 Linux web servers in it. (running version 10.2.1) Two of the webservers are physical machines, and two are virtual.
I have selected round-robin as the load-balancing method, disabled priority group activation, and the ratios for each member in the group is set to 1, connection limit for each member is set to zero.
Whenever I look at the statistics, the physical servers are getting about 6 to 8 times the connections of the virtual machines.
The physical machines do have more memory then the virtual machines, but I've verified the virtual machines have plenty of free memory available, and the load average is roughly the same across all the servers.
So I have no idea why the Virtual Machines have such a low connection count as compared to the physical machines. I've checked the apache configs and made sure the number of servers, threads, etc are set to be the same across all machines.
Any clues or things I should check for to verify the round-robin is working as expected, or is this a bug or expected behavior?
Thanks,
21 Replies
Taylor77_133376Sys::Hardware Chassis Fan Status Index Status Fan Speed(rpm) 1 up 9926 2 up 10546 3 up 9926 Chassis Information Maximum MAC Count 2 Registration Key - Chassis Power Supply Status Index Status 1 up 2 up Chassis Temperature Status Index Temp(degC) Location 1 36 Air Inlet CPU Status Index Temp(degC) Fan Speed(rpm) 1 46 10546 Hardware Version Information Name cn0 Type crypto Model Cavium NITROX-PX Parameters -- -- version CNPx-MC-SSL-MAIN-0013 Name cpld Type pic Model F5 cpld Parameters -- -- version 0x13 Name cpus Type base-board Model Intel(R) Pentium(R) Dual CPU E2160 @ 1.80GHz Parameters -- -- cache size 1024 KB cores 2 (cores/cpu:2) cpu MHz 1800.108 Platform Name BIG-IP 1600 BIOS Revision F5 Platform: C102 OBJ-0332-01 BIOS (build: 130) Date: 12/28/09 System Information Type C102
Joe_MWhat is the uptime on the device?- Taylor77_133376228 days
- Joe_MThere are some known issues with uptime. Non with round robin that I am familiar with but doing a reboot wouldn't hurt. Also, at some point, I would definitely recommend upgrading to 10.2.4 with the latest Hotfix.
What_Lies_Bene1Cirrostratus
OK, thanks, so it's a CMP system. Are we talking about a low number of connections or not?
- What_Lies_Bene1
OK, so the system distributes connections across the two TMMs running (one per core) so you'd be down to 75/100 per TMM. Even if one started the Pool RR allocation at Pool Member 1 and the other at 3 you should still end up with a fairly even spread I'd say. That being the case I'm a bit stuck.
Kevin_Davies_40Nacreous
Are the pool members 37/38/39/42 all separate Linux servers? Some on hardware and some vmware?
If you watch a single connection serverside with tcpdump to the hardware version and do the same with a connection the vmware server is the behaviour of the connection the same?
- Taylor77_133376Yes, seperate linux servers, 2 physical, 2 virtual. TCPdump looks about the same. One thing I notice, is from the f5, pings to the physical boxes take about half the time (averaged over several pings) as compared to pings to the virtual servers.
