RE: [load balancing] switch redundancyI Think that most IT specialist, also in this LB area are focused on PCs and workstations and their capabilities.
In a mainframe environment, servers are quite different in their IP behaviour.
Lesson one - An OSA-Express adapter in a mainframe is a dual-homed adapter. Typically, a site has _LOTS_ of these.
Each adapter has 2 physical interfaces. Both Active.
You can put these 'interfaces' in the same subnet, or not. Your choice. This is not the actual problem.
A Mainframe also has a concept of a VIPA address. This is not the same as the address of the interface addresses.
It can be in a completely different subnet and mask as the real interface address.
So you have at least a hop from the interface address to the actual application VIPA address.
This makes a problem for load-balancing devices that would like to be able to be active-active.
So, having two active CSS load balancers in a multi-homed mainframe environment, is _NOT_ possible. Period.
At least Until they do flow-sync on flight, or do like the MNLB solution from Cisco.
----- Oprindelig meddelelse -----
Fra: Kenneth Schow
Sendt: 20. august 2001 20:51
Emne: RE: [load balancing] switch redundancy
The switch redundancy problem depends on the capabilities of the OS and the load balancers. There are usually two stipulations to the dual redundant switching solution. First, you must be able to dynamically bridge two NIC cards on the server, and correctly accept gratuitous arps. Second, there needs to be a catalyst from the load balancer, such as a gratuitous arp in the event of a fail over. Typically, most unixes are able to accomplish the first caveat through bridging at the kernel level (or multipathing on Solaris). The catalyst depends on the design of the load balancer. I know most LB products send arp updates when they fail over, in this case HSRP does as well (rfc2281). Therefore, if you bridge the NICs, and they can accept gratuitous arps. The expected result would show your arp table being updated, and traffic should pass uninterrupted.
From: Patrick McAndrew [mailto:PMcAndrewIZZATstudentadvantage.com]
Sent: Monday, August 20, 2001 10:48 AM
Subject: RE: [load balancing] switch redundancy
The two nic's will be connected to two layer two switches. The switches will
be connected to a pair of load balancers. The servers will be in a seperate
vlan, their network will be a RFC1918 compliant one.
I am assuming the server nic's would be in seperate bridge groups?
I am going to be running HSRP on the switches, but I'm wondering how to get
the server to notice when a switch is down, and use the other interface.
Does anyone have experience with doing that? If I don't make myself clear,
> -----Original Message-----
> From: Stephane LITKOWSKI [mailto:st_litkowskiIZZAThotmail.com]
> Sent: Monday, August 20, 2001 11:48 AM
> To: lb-lIZZATvegan.net
> Subject: Re: [load balancing] switch redundancy
> What r u talking about ? Layer 2 switches ? or Layer 3 or 4 switches ?
> Are the server nic's bridged between them ?
> ----- Original Message -----
> From: "Patrick McAndrew" <PMcAndrewIZZATstudentadvantage.com>
> To: <lb-lIZZATvegan.net>
> Sent: Monday, August 20, 2001 5:29 PM
> Subject: [load balancing] switch redundancy
> > Hello all,
> > I'm wondering how some of you use switch redundancy.
> > Here's the situation:
> > I want my webservers to have two nic's in each box.
> > The two nic's will be connected to two differant switches.
> > They will have an identical configuration, I don't want to
> do anything
> > fancy, just provibe switch redundancy.
> > What is a good way to go about doing this? STP? How will
> the servers know
> > not to use one nic as opposed to the other?
> > Thanks in advance!
> > Pat
This archive was generated by hypermail 2b30 : Tue Aug 21 2001 - 22:12:41 EDT