Hi there, I am thinking about active/active LTM deployment which each LTM is located on different data center and using different subnet. I mean not using GTM at all. For example data center A - subnet A data center B - subnet B I think if I create VIP on subnet C and D and then configure upstream router to route traffic of subnet C and D to appropriate data center e.g. dynamic route, static route, etc. If one of data center fails, upstream router will be notified and move traffic accordingly. Is this possible? What do you think? Has anyone used this configuration? Thanks a lot! Nitass

There are a number of reasons why you wouldn't want to do this/wouldn't work.. but I'll touch on the most obvious. Without any intelligent DNS, aka GTM, how would you handle DNS? With the boxes being on different subnets, your VIPs are going to be different... Active/Active really wasn't designed to work like that..For something like that to work you would have to bridge the DC's.. very bad... I can keep going here but you shouldn't focus on this. The recommended setup by F5 is Active/Standby. You will need two Boxes in each DC. There are a number of reasons why you would want this setup, redundancy, change control, etc. If you want global redundancy between DC's you are going to need additional boxes/licenses for GTMs to control this.

I might be wrong. Anyway, I think VIP will not be changed. For example, VIP C is hosted on data center A and VIP D is hosted on data center B. In normal condition, upstream router will route C subnet to data center A (physical subnet A) and D subnet to data center B (physical subnet B). However, if data center A fails, VIP C will be moved to data center B and upstream router has to be modified somehow to route C subnet to data center B. Is this possible? Please let me know if I missed anything here. Much appreciate!

The high level answer to your question is Yes, but the devil is in the details. The GTM is to Datacenter Load Balancing what the LTM is to Local Load Balancing. The GTM can do Active / Active Load Balancing based on whatever criteria you provide like Geographical Location. However, the VIP's are not "moved", the traffic is redirected at the GTM Level based on Health Checks that the GTM does on the Local Datacenter LTM's (much the same way that an LTM marks servers available or unavailable in a Pool). I am seriously over simplifying the setup description, but it this is something you are interested in then you will need to do additional research. Keep in mind that the F5's / BigIP's do not do data concurrency for your applications (you have to do that on the back end) and that alone might kill the concept of Active / Active.

Remember in his first post he said no GTMs... Two things will help us help you.. Can you please post a drawing of what your current setup is, stripping any private info..? Can u also state what you're looking to accomplish? Leave out how u think it is technically going to work..a high level analysis of what u have and what u want to accomplish will do.. Thanks!

Thanks you both! Yes, this is no GTM. Actually, it is just an idea I think it could be possible. Anyway, I have not had a chance to test it in lab (and it may not be applicable for testing since lab has limited resource). So, I would like to know if there is anybody having done it before or having something I missed.

active/active LTM deployment on different subnet

16 Replies

The_Bhattman
Nimbostratus
Mar 18, 2011
HI Nitass,

I believe what you are asking is possible. I have lab tested something similar several years ago where I wanted to avoid using a GTM or like service and found that it would increase complexity across the board and it was not fully automated - some manual manipulation was required.

My lab setup was the following

Datacenter A

Circuit A

VIP A (Active)

VIP B (Dormant)

Servers A

Datacenter B

Circuit B

VIP B (Active)

VIP A (Dormant)

Server B

I ran BGP peering between Datacenter A and B (which used 2 different ISPs). I created VIP A and B on Datacenter A and the same in Datacenter B. However, only the VIPs that were being advertised in their respective ISPs were ever responding. I then created a route pools which custom health checked IP addresses on the upstream router (checking for latency and ICMP). Since ADCs in both data centers were connected to switch routers (Cisco 6509s), I created secondary addresses on the subnets (VIP A and B).

This worked well in the lab, however, I found one large issue. Because BGP was being used, fail over could take anywhere between 30 seconds to 5 minutes depending on your ISPs defaults. I even looked into beefing up BGP with PfR(Performance Routing). However, this required the ISP to support it and not all ISPs offer this type of configuration. I even looked at lowering the BGP timers - again the ISP had to support it on their end which is highly unlikely when you deviate from the known standards.

In conclusion, I decided against this approach because it simply could not guarantee consistent failover within a reasonable amount of time as well as the complexity in maintaining this configuration. Thus we decide that the GTM would provide the best solution overall.

I hope this helps

Bhattman
nitass
Employee
Mar 19, 2011
Hi Bhattman,

Thanks!

/Nitass
Austin_Geraci
MVP
Mar 19, 2011
That's the catch, even if u use a flavor of peering there is a lag in failover.. even if it's internal and u can control it... still not the right way to do it..
nitass
Employee
Mar 19, 2011
Hi iRuleYou,

Thanks again! :-)

/Nitass
EIT_Network
Altostratus
Dec 23, 2019
Hi Nitaas,

i am stucked with the same scenario as you were in that time. Only difference is that we have Seperate load Balancer pair for Second DC.

Clients has some sites on DC-1 and some applications hosted on DC-2. We can say this as Active-Active Data Center (in LB prospects). For GTM - We decided to play with Listener IPs (on both data center) on Public DNS for CNAME entry. for Eg:

Listener IP DC-1 : 10.10.10.1
Listener IP DC-2 : 10.20.20.1
Application -1 : www.xyz.com (at DC-2)
Application-2 : www.123.com (at DC-1)

We'll do that CNAME entry/Delegation record entry on Public DNS with both DCs listener IPs for both the applications.

i.e. www.xyz.com 10.10.10.1
10.20.20.1

www.123.com 10.10.10.1
10.20.20.1

and Set the member order/ratio on for each application at respective DCs accordingly in GSLB wide IPs members.

-> 0 for application-1 at DC-2

-> 0 for application-2 at DC-1

Assuming Public DNS is performing Round Robin load balancing only.

highlighting the important point : VIP IP pool and Server IP pool would be different at these data centers.

Could you please help us to about the approach you adapted or Else pls sugget for other effective solution.
- nitass
  Employee
  Jan 06, 2020
  Sorry for delay. I am a bit busy lately. I do not remember what it ended up to be. 😅 It has been a while. Let me dig it and will let you know.

Forum Discussion

active/active LTM deployment on different subnet

16 Replies

Recent Discussions

Disk space full - what files, folders are safe to delete?

ASM instance creation

BIG-IP DNS: Check Status Of Multiple Monitors Against Pool Member

Open Redirection Mitigation

F5Access | MacOS Sonoma

Related Content

Active/Active load balancing examples with F5 BIG-IP and Azure load balancer

Where are F5's archived deployment guides?

F5 BIG-IP deployment with OpenShift - per-application 2-tier deployments

BIG-IP L2 Virtual Wire LACP Passthrough Deployment with IXIA Bypass Switch and Network Packet Broker (Single Service Chain - Active / Active)

Azure Active Directory and BIG-IP APM Integration