Forum Discussion
chris_connell_1
Nimbostratus
Nimbostratusraito load balancing not working
Hi,
We are seeing a weird issue with ratio load balancing on ltm 10.2 Have tried both node and member methods, same result. Basically the connections are not weighted anywhere near what we have set. I am pretty sure it was working in version 9.x but not sure. We are not using any features like one connect or priority groups but we are using source address persistency. Any help appreciated.
POOL MEMBER TEST_POOL/10.12.0.202:any active,up
session enabled priority 0 ratio 2
(cur, max, limit, tot) = (2851, 3515, 0, 338467)
(pkts,bits) in = (9.665M, 8.587G), out = (15.50M, 148.4G)
requests (total) = 0
+-> POOL MEMBER TEST_POOL/10.12.0.203:any active,up
session enabled priority 0 ratio 1
(cur, max, limit, tot) = (1260, 1609, 0, 141516)
(pkts,bits) in = (3.832M, 4.056G), out = (6.009M, 56.45G)
requests (total) = 0
+-> POOL MEMBER TEST_POOL/10.12.0.204:any active,up
session enabled priority 0 ratio 3
(cur, max, limit, tot) = (2106, 2686, 0, 236941)
(pkts,bits) in = (7.619M, 7.214G), out = (12.60M, 122.9G)
requests (total) = 0
- Chris_Miller
Altostratus
Persistence will certainly have an effect. Are clients being NATed at all before they hit the VIP or is original source IP preserved? 
chris_connell_1
The clients are not being nated, their source IP is preserved. Could you help me understand why persistency will have an effect? I see one reason, which I guess why you asked about NAT, is that if a bunch of connections originated from the same source IP address. This should not be the case as we have the mask set at /32 and the clients have internal addresses, but I will double check.
natheCirrocumulus
You're right in your reason re NAT - CM may have more reasons too.
Anyway, what's your Persistence Timeout set to? Could this be affecting the load balancing of the connections - are connections based on persistence counted in the stats above? Not too sure myself without checking.
Anyway, are you able to clear the persistence tables and see them build up to see how the initial ratio weightings are done. If persistence is affecting things then you'd guess initially the load balancing would look right.
N
Trosho_94229Hi,
Any other ideas on this case?
I still can not understand how persistence can affect ratio load balancing ? Or my misunderstanding is in the way the Ratio LB works.
It chooses next node for next connection where connection is each TCP session or new client IP ?
Regards,
Todor- Chris_MillerTrosho - before a load balancing decision is made, LTM will check its persistence table to see whether the client has an existing session. The Load balancing algorithm, best it least connections, ratio, round robin, etc, only affects new connections. Make sense?
- Trosho_94229Hey Chris,
This is my understanding too.
What''s wrong in the below:
Let's say we have two nodes in the pool with ratio (member) LB and ratios 2 for the first and 1 for the second.
We have persistence set on src_ip.
3 different clients IPs.
1st client makes connection (TCP session) - 1st node. 1st. client makes second TCP session - again 1st node because of the persistence.
2nd client makes connection (TCP session) - 1st node (because of the ratio)
3th client opens TCP session - 2nd node
T - Chris_MillerTrosho - The behavior is a bit different than that. Your 2nd client will actually go to node 2 and your 3rd client will go to node 3. It's basically round robin until each member has a connection.
The LTM Essentials covers this behavior and is free on the web.
http://www.f5.com/services/global-training/web-based-training/ - Trosho_94229The order is not of big importance.
How the above can happen if we have persistent on source IP and time of IP address reassign is much grater than persist timeout
T - Chris_MillerDo you have other persistence options? Can you use cookie for instance? Is this an HTTP Application?
- Trosho_94229I have persist over virtuals and persist over services.
There is one pool and few virtual servers.
There is HTTP but not only and we can't rely on cookies.
Regards
