about ip_strong_es_model and lan fail over/redundancy

[Xiang Zhao_1]

Hi all,

The customer use Alteon Layer4 switch with several HP PR3440 to provide web service. Each RP3440 has multiple lan cards.

During the past several years, we configured two IPs in the same subnet on these two lan cards , and make each IP as a real server. If one link is down, the layer 4 switch at least can use another real server(IP), without puting the whole server down.

However, we finally found out since all the out-bound traffic goes to the same lan card(with ip_strong_es_model default to 0), if that link is not functional(and the operating system didn't detect that problem, still shows UP status), the whole server(two real servers) is down. Generally, we have to manually disable the lan card with the bad link to restore the connectivity of another link.

Here I'd like to know without using APA(Auto port aggregation), MC/Service Guard, how to solve this redundancy problem?

One option is to only activate one lan card, fail over to another in case of failure. I know we can write scripts to monitor the link status and do fail over by ourselves, it this practical ? Anyone has sample scripts to do this?

Another option is to activate two lan card simultaneously, configuring them in one subnet or two.
If in one subnet, the network equipment configuration is simple, which should also lead to easy maintenance. However, I'd like to know whether HP-UX supports multiple IP in the same subnet? and our RP3440(hpux 11.11) shows ndd -h ip_strong_es_model only have two value, 0 or 1. Only after installing Transport Optional Upgrade Release (TOUR), can we use value 2? Also anyone explain exactly what the following means (excerpt from the TOUR 2.4 release notes (http://docs.hp.com/en/5991-0782/5991-0782.pdf)

"If ip_strong_es_model is set to 1, the strong end-system model is enabled on the system.
Starting with TOUR 2.4, the ip_strong_es_model parameter supports another value, that is 2. If you set ip_strong_es_model to 2, the system attempts to match the best gateway
(discussed in Case 1) when multiple default gateways are set up on a single IPv4 subnet.
Following are different instances that affect the routing decision:
Case 1 If the client application is bound to the IP source address, the default gateway
chosen is the one associated to that source address.
Case 2 If the client application is bound to INADDR_ANY, the default gateway chosen is the
one associated to the source address the client application binds to."

From Rick Jones message(http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1015399), value 2 means "source IP is included in route selection, without the restriction on inbound IP datagrams being accepted only on the interface with that IP assigned", However I can't derive this explanation from the TOUR 2.4 release. Also Rick Jones didn't mention the TOUR, does the latest HP-UX 11iv1 supports value 2 now without installing TOUR?

Even we configure these two lan cards in two subnets, we're still facing the multiple default gateway problem, so we still need to deal with ip_strong_es_model option? and do we need to install TOUR? Does HP-UX supports "multiple default gateway" offically? If yes, how? do we need to install TOUR?

BTW, in this environment, performance(throughput) is not an issue, we focus on redundancy.

Or there are any other ways to solve this redundancy issue in this environment(I mean Layer-4 swich with HP hosts) without APA and MC/SG?

Maybe too many questions, Thank you all in advance for your kind help.

-Xiang

P.S:
the reason not considering APA: APA is not free, and APA use virtual lan card(by my understanding), configuration seems complex.

the reason not considering MC/Service guard: MC/Service guard even more expensive than APA, and configuration is complex.(we need to setup a single node cluster, package,etc.).

[rick jones]

ip_strong_es_model is only in the latest TOUR(s). ISTR a 3.X TOUR but that is fuzzy. I believe that includes 11iv2 but I've been out of the UX loop long enough now that I may have that wrong. One would expect ip_strong_es_model == 2 to be part of 11iv3 when that starts shipping.

ip_strong_es_model of two basically uses the outbound mechanisms of ip_strong_es_model == 1 and the inbound of ip_strong_es_model == 0. That is, it will try to send IP datagrams out an interface based on both source and destination IP, but will accept datagrams on any interface.

If the failure mode of the setup is such that the link on the system actually remains UP, I don't think that ip_strong_es_model == 2 will actually help anything.

If a link between the UX system and the switch is _really_ not functional, the UX side is _supposed_ to be detecting that, and not detecting that would be a bug which needs to be brought to the attention of the Response Centre.

Now, if the link in question is really between a switch and the load balancer, (ie not directly connected to a NIC in the UX host) then the link-layer stuff by definition cannot know that link is down.

[Xiang Zhao_1]

Rick, thank you for your prompt answer.
Let's forget the hp-ux network connection detection problem, just as you mentioned, suppose there are switches between the hp box and the load balancer.(pls kindly see the attached topology)
and suppose we configure the two lan cards(lan1, lan2) in the same subnet. If ip_strong_es_model is default to 0, then the out-bound packet will always goes to one lan card(suppose it's lan1), then if the link between switch1 and load balancer 1 is down(red line in the attached topology), this will also affect the connectivity of lan2. So the whole server(web1) will be down. If ip_strong_es_model is 1, then the hp-ux will send reply package to the client via lan2 when the load balancer access real server2(lan2). It has nothing to do with lan1, thus web1 is still functional(via lan2, real server 2).
I don't know the above understanding is correct or not, if it's correct, ip_strong_es_model will solve my redundancy problem.

However, even it works, I'd like to know whether HP-UX 11iv1 supports ip_strong_es_model OFFICIALLY, that is, supports "two IP address in the same subnet". And from my understanding, I have no need to install TOUR to have value 2 functionality. (value 2 might help if we need to have two real server up even one link is actually down, which is not necessary in our scenario).

And if "two ip address in the same subnet" is not a good practice, we might need to configure them in two subnets. However, even in two subnets, we still need to setup multiple default gateways, and we need to install TOUR, which allows us to configure a default gateway for each IPv4 interface. I also like to know my above understanding is correct or not.

Or again, is there any other solution to meet my redundancy requirement without APA and MC/SG?

Thanks in advance.

-Xiang

[Xiang Zhao_1]

anyone can help me? thanks in advance.