_______________________________________
From: Jose Hernandez [mailto:[email protected]] 
Sent: June 12, 2009 5:29 AM
To: [email protected]
Subject: [pfSense Support] Inbound load balancer performance under heavy
load.

Hi,

Yesterday we had a service launch, and pfSense inbound load balancer let me
down big time... We have been using pfSense 1.2-release version installed on
Dell PowerEdge R200 and CARP for redundancy for around a year now, it probed
to work although we never have had a very high load.

Yesterday right after we launch the service, we start getting complaints of
many requests failing from users. After some investigation it was clear that
the request were not getting through to our systems!!!

The only indication of something going bad was the traffic graph (attached
is a screen grab), it was picking up and down as never before... We did some
load testing last week and the week before and we were seeing ~100Mbps
constant outbound speed, we also have seen in the past ~100Mbps inbound
speeds... So I first blame our IP transit provider, after contacting them,
they confirmed to me that no packets were being lost or dropped anywhere in
their network and that their systems were just fine... so the only other
thing that could be causing the problem was pfSense... however I couldn't
find any indication of anything going wrong but the traffic graph... memory
and processor were fine, states table size, no packets dropped in RRD
Graphs, etc...

After tweaking many settings in pfSense with no joy, I finally removed the
Virtual Server and created a NAT Port Forward to only one of our web servers
layer at the backend... and that fixed the problem of requests not getting
through and the traffic graph was again stable... I wonder if it is there
any known issue with the inbound load balancer... I think the problem was
with the number of source IPs or states it had to deal with (after the load
balancer was removed, the states picked up to ~210000, as when load testing
we tested from a bunch of ~10 IPs... 

The problem is that we do need load balancing, mainly for redundancy of our
systems at the back end...

The inbound load balancer that was set up had 3 servers in the pool and, the
port was HTTPS and TCP monitor was configured

Is there anything in version 1.2-release that affects the performance of the
inbound load balancer? Would this performance issues go away if I upgrade to
the latest stable version, currently 1.2.2?

We are also thinking in getting commercial support, however we are not sure
if this will help as we don't know if pfSense is actually able to take the
load...

Can anyone shed some light into this issues we are having?

Regards,

Jose Hernandez
Software and Systems Senior Engineer
VIDZONE DIGITAL MEDIA
 
GET IN THE VIDZONE(tm)
 

What about cpu load at the time the problem was happening? Do you use smp
kernel?
http://forum.pfsense.org/index.php/topic,6805.0.html

Eugene


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Commercial support available - https://portal.pfsense.org


Hi Eugene,

I have attached RRD Graph for the processor load, the problem started at on
around 1500 and it was around 1830 when I removed the load balancer. The
processor utilization would be higher between 1830 and 2300 because at those
time I had the Firewall Optimization Options set up to 'conservative'... I
tried that as the problem was packets lost... but as soon as I changed back
to 'normal' the processor utilization dropped as expected.

Regards,
Jose



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Commercial support available - https://portal.pfsense.org

Reply via email to