IIS 7 & IIS 8
Step to achive 2tier WebFarmFramework with ARR including HA using NLB
Last post Jan 22, 2013 07:39 PM by znerses
Nov 15, 2012 04:38 PM|LINK
Thank you for your reply. Could you perhaps specificy option 2 some more? I've tried this setup, but I didn't have any success due to the fact that the 2 WFF servers were trying to manage 1 (the same) Primary Node... Last time I tried i used WFF 2.1 and ARR
on both 'controller' servers.
Dec 05, 2012 09:28 PM|LINK
I had a few questions and was hoping that you or someone here who is more knowledgable about ARR could assist me. We currently have about 10 standalone web servers hosting an application. The 10 web servers are tied to an NLB address that provides network
load balancing. We would like to implement 2 ARRs (using HA and Load Balancing) in our environment to intelligently route web requests. I read the step-by-step instructions in your first post but i am having trouble understanding the setup of ARRs.
1. Is it possible to install ARR without putting our Web servers into a Farm? What are my different options? We don't want to screw up our current Web servers.
2. Do we need to put 2 ARRs into their own Farm? and web servers into their own separate Farm?
3. What would happen to our current Web servers if we put them into a Farm?
4. If the Primary web server fails, can we still update the Secondary servers? What would happen if the Primary Web server in the farm goes down?
5. Will ARR still be able to function if one of the Controllers goes down? I am assuming the 2nd ARR will read the "shared config" in the offline files?
6. "Make sure ARR02 can reach the shared config folder whilst ARR01 is offline! If not, make it so." How do I check this?
7. "Create a server farm and add nodes, including a primairy node." Where to do this? on the 1st ARR or another web server?
Please let me know what is the best way to go about starting my setup.
Dec 10, 2012 10:06 AM|LINK
My setup that announced here earlier is now fully functionnal :
2 WFF+ARR controllers configured in NLB (Active-Passive for the moment). They use shared configuration for their XML config files located on a network share.
Additionally that network share is configured in Offline Files on the two controllers to avoid problems of share unavailability. Without this I experienced some glitches as reported here previously
Each weekend, I restart webfarm controller service on the controller servers to avoid memory leaks problems observed in devellopment environment.
No problem (request fail) reported since almost two month now.
Concerning use or not of ARR, I totally agree with LukaszH. But to my opinion NLB load balancing alone is not efficient emough. ARR consists in writing rules of traffic routing based on many different variables what is much more powerfull and without ARR,
forget about multiple web farms.
Jan 22, 2013 04:14 PM|LINK
It seems to me that both yvanm and LukaszH when they use their 2 x (WFF+ARR) in NLB + 1 primary and x secondary nodes setup, they have only one website running there or at least one web IP address that clients request. Is that correct???
In my case I am looking to have 60-80 websites in that system and most of them require their own dedicated IP address due to SSL and FTP needs.
In that case would it still work with the same setup?
Client request --> ARR's NLB IP--> ARR 1 or 2->Nodexxx
In my case clients will be asking for a different IP address each time.
Jan 22, 2013 07:17 PM|LINK
Well, i think it might not run that easily
This is mostly dependant on the max capacity of NLB (The amount of loadbanced IP's). NLB is sweet, but It's a pretty old feature and perhaps it might not support loadbalancing those quantities. Regardless, you can always go for other load balancing software
or, even better, hardware.
Then, i must say that my experience with ARR + WFF has had freakishly high memory usages. So, suppose your websites have many application changes. ARR will notify and sync all nodes, which uses al lot of memory, as well as network traffic, and might eat
unacceptable amounts of RAM.
Also, you might also run into the next problem; NLB on the same hardware as the ARR, which so many hosted websites, i can imageine you have quite a bit of network traffic going through this hardware. The amount
of concurrent connections giong through this hardware could be a bottleneck.
These are some thoughts that need testing. I must state that my experiences are with Server 2008 R2 + WFF 2.1 (including some hotfixes) and i've read that Server 2012 packages a whole new version of IIS and the Web Farm Framework , which, i can imagine might
bring the WFF to a new level. I suppose it could be done with WFF2.1, when you have a lot of static applications / websites and low networking traffic.
Please keep in touch about your findings.
Jan 22, 2013 07:39 PM|LINK
Hi Sam, thanks. I will keep you posted on how that goes. Luckily the outside traffic is not that high, so it might work, but I do hear you on internal traffic and memory usage, this is something I will have to monitor closely and see if it works.
I am more leaning towards doing just WFF instead of WFF+ARR. This way it will be following this flow:
Client request--> NLB IP between nodes-->Nodexxx (with controllers watching over; I still intend to have two controllers in a fail-over cluster (Active-Pasive)).
This is a simpler set-up, but I still have some concerns as follows:
1. if controllers go down, will clients still be able to access the sites?
2. With this set-up it means every time I add a new site with a new IP address, NLB is going to get unstable. It seems that any time I make a change to NLB configuration (like add IP address, add port rule, change load %) it completely freezes, because instead
of copying these changes to other nodes it just notices configuration dioference between NLB nodes. Does anyone else experiance this? I read a MS article on it with a hotfix, but the hotfix didn't do anything.