Forum Discussion
2funky_105078
Cirrus
CirrusGTM: avoiding flapping DNS answers with RTT method
I am in the need to understand how GTM metrics work for GTM LDNS probes.
1) how they can be logged the decision? I am using 11.2 but moving fast to 11.4.1.. :)
2) Lets make an example. if our GTM chooses a VIP in USA for 100 consecutive times because the RTT is lower going to this USA VIP and then for 1 time it gets a better value - for whatever reason - to go to another VIP, for example to Australia, will it be considered valid the last value which differs from the original 100 previous time? is there cache variation value that can be configured to avoid this flapping choices? (We had this choice in Cisco GSS).
3) how long is the non-optimizes Australian value kept in cache until a new value is reconsidered? It is the Inactive timeout of 28 days?
The only solution I could see in your situation is to break your single pool into three pools, and either stick with RTT and enable persistence at the wide IP, or switch to topology. Either one should get you some more consistency.
2funky_105078Thanks Cory i think i have no more questions now :)
Cory_50405Noctilucent
The only solution I could see in your situation is to break your single pool into three pools, and either stick with RTT and enable persistence at the wide IP, or switch to topology. Either one should get you some more consistency.
- 2funky_105078
The problem - differently said - is that if i have only 3 VIPs associated to a global application and this should be accessible from anywhere in the world, there will always be a place where there is flapping.
For example, if i have a user with LDNS exactly in the middle (from the RTT point of view) between 2 different DCs, how can I remove the tedious flappping issue?
- Cory_50405
That's correct. You'd need multiple pools and multiple pool members in each pool to fully utilize the topology/RTT load balancing setup. Knowing your current setup, I see why you chose RTT vice topology.
- 2funky_105078
Well, let me see if I understand it with an example.
Right now, I have configured a wide-IP (like www.example.com) with just 1 Pool associated to it, consisting of 3 Memebers/VIPs: 1 in USA, 1 in EU and 1 in AU.
If I got it right, you are suggesting that I could configure 3 Pools associated to this Wide-IP and use using Topology LB to choose among these 3 Pools . But, then in each Pool there is only 1 VIP/Member, so I cannot really use RTT with just 1 VIP inside.
Maybe your case is applicable in case we have multiple Members within a Pool, right?
- Cory_50405
You can apply topology based load balancing at the wide IP level to select a pool. You can then apply RTT load balancing at the pool level to select a member. So they can be used in conjunction depending on your setup.
If your pools are built according to data center, then topology could send users from a certain country/state/etc. to a certain pool. The pool could then use RTT to send you to the lowest latency pool member.
In your example, I don't see a way for GTM to accomplish what you wish to do.
- 2funky_105078
Thanks Cory.
Can topologic based load balancing work in conjunction with RTT, or it is either one or the other way?
Can we use standard RTT and "correct" the anomaly values with the information coming from the topology?
For exmaple, if we get `back USA DC VIP for 100 times and for just 1 time we get Australia VIP because the RTT came lower there just one time, then can we ask GTM to have a look also at topology based info to check if it really needs to flap to another DC VIP?
- Cory_50405
RTT load balancing is always going to return the answer with the lowest RTT independent of previous results. The amount of time that an answer will remain in an LDNS cache will depend on the TTL you are setting on the GTM. That is if the LDNS respects the TTL, which some do not.
If you are looking for more reliability with regards to resolving DNS queries to data centers/servers/etc. within a certain region, you may want to explore topology based load balancing. You can use it to direct users from countries/regions/states to data centers/pools/etc based on geographic location.
http://support.f5.com/kb/en-us/products/big-ip_gtm/manuals/product/gtm_config_guide_10_1/gtm_topology_newest.html