04-03-2007 12:55 AM
Hi, From the Cisco pages I read the following
"The CSS uses the weight for a service when you configure weighted roundrobin load balancing on the content rule. When you assign a higher weight to the service, the CSS redirects more requests to the service."
This suggests to me a server with a weight of 2 will receive twice as much hits as one with a weight of 1.
We have this configuration but the customer sees the server with a weight of 1 getting more hits, this is backed up by the show summary stats on the CSS.
content www.abc.com
vip address 145.26.114.42
port 443
protocol tcp
advanced-balance sticky-srcip
sticky-inact-timeout 31
add service 1.1.1.1-443
add service 2.2.2.2-443 weight 2
add service 3.3.3.3-443 weight 2
add service 4.4.4.4-443 weight 2
add service 5.5.5.5-443 weight 2
add service 6.6.6.6-443 weight 2
add service 7.7.7.7-443 weight 2
balance weightedrr
active
=======================================
Service Hits
2.2.2.2-443 2784623
3.3.3.3-443 2868285
4.4.4.4-443 2688701
5.5.5.5-443 2763660
1.1.1.1-443 5105735
6.6.6.6-44 2634987
7.7.7.7-44 2733737
cheers,
Mike
04-03-2007 04:11 AM
Mike,
the sticky function will overpass the weighted roundrobin.
So, if the first user open 1 connections with server 1 and the 2nd user open 10 connections with server 2, you will see unequal load on your servers.
This is due to stickyness.
Gilles.
04-04-2007 12:22 AM
Hi gdufour,
I have the same situation, weightedrr + L3-sticky. if the range is 0 to 10, does it mean 2 weighted more than 1? but however users stick to service with weight 1 tend to open more connections than users stick to service with weight 2, so problem mentioned arise, is that the cause?
I'm still confusing on how to calculate this.
Regards,
Hao Dai.
04-04-2007 12:44 AM
Gilles,
I can't really believe this is the cause, prior to configuring the weights the hits on the box were almost identical (+/- 5%).
Now the server with weight of 1, is almost 50% more than all the other 6 servers, if stickyness was the cause then each of the 7 servers would have differing hits, 6 of the servers are extremely close in hits.
The nature of this application would not cause the a user to open one connection, and another to open 10.
cheers,
Mike
04-04-2007 06:40 PM
Hi Mike,
I consulted Cisco SE and TAC engineer, they confirmed that higher weight number brings more conns to service. When changing the sticky-mask, it seems to me the sticky-table is updated instantly and conns lay on services by the ratio it meant to be. It works for me on production network anyway, maybe you want to try it if you've got a lab environment.
Regards,
Hao Dai
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide