Alternative network setup

z/OS network adapter

On the z/OS side this, configuration uses Address Resolution Protocol (ARP) takeover to transparently handle the failure of one OSA-Express feature in a redundant setup of at least two physical interfaces.

Usage of ARP takeover: ARP takeover is a function that allows traffic to be redirected from a failing OSA connection to another OSA connection. This function is supported by IPv4 and IPv6 OSA interfaces. ARP takeover uses LAN broadcast protocols. Therefore, to use ARP takeover, all of the z/OS systems that belong to your SAP installation must be in the same LAN respectively in the same TCP/IP subnet. This is called a flat network topology. No bridging, no routing, all participants are within the same subnet. With ARP takeover, you can achieve availability that is to a high degree comparable to what can be achieved using dynamic routing.

Dependencies and restrictions of ARP takeover: ARP recovery solutions rely on the fact that hosts accept a new MAC address for a given IP on the same network medium. However, this has a major disadvantage in that other media such as HiperSockets or XCF cannot be used as alternate routes.

• ARP takeover requires dedicated OSA-Express adapters (ports) in QDIO mode.
• If you want to move an IP address from one adapter to another within a system, you must also set up a Static Virtual IP Address (VIPA).
• While it is possible to configure and use multiple parallel paths to increase available bandwidth (called multi-pathing) without a dynamic routing protocol, if there is a failure in the network beyond a point where the OSA adapter is able to detect it, TCP connections that are directed across the failed path will time out and UDP and RAW traffic is lost.

For more information, basics and configuration samples on ARP high availability, refer to z Systems OSA-Express Customer's Guide and Reference.

z/OS TCP/IP configuration sample for the z/OS 1 LPAR (extract): The static VIPA of LPAR z/OS 1 is 10.101.6.188 and the SAP application talks to this VIPA only. The VIPA is mapped onto the real interface via the SOURCEVIPAINTERFACE statement, which makes the real interface IP transparent to the application. The z/OS 1 static VIPA 10.101.6.188 can be active either on the OSA interface with 10.101.6.140 or 10.101.6.141. The ARP message sent out is the one of the real interfaces.

DEVICE  VLINK2   VIRTUAL  0         ; VIPA
LINK    VLINK2   VIRTUAL  0 VLINK2  ; VIPA
;
INTERFACE SYS1IF1                   ; INTERFACE TO SWITCH 1
DEFINE IPAQENET
IPADDR 10.101.6.140/26
PORTNAME SYS1P1
SOURCEVIPAINTERFACE VLINK2
MTU 8992
NONROUTER
;
INTERFACE SYS1IF2                   ; INTERFACE TO SWITCH 2
DEFINE IPAQENET
IPADDR 10.101.6.141/26
PORTNAME SYS1E2
SOURCEVIPAINTERFACE VLINK2
MTU 8992
NONROUTER
;
HOME
  10.101.6.188             VLINK2   ; VIPA
;
 BEGINROUTES
; NETWORK          MASK   FIRST HOP    LINK          PCKTSZ
  ROUTE 10.101.6.128/26   =            SYS1IF1   MTU 8992
  ROUTE 10.101.6.128/26   =            SYS1IF2   MTU 8992
ENDROUTES
;
  START SYS1IF1                     ; INTERFACE TO SWITCH1
  START SYS1IF2                     ; INTERFACE TO SWITCH2
;
 BSDROUTINGPARMS TRUE
  VLINK2 8992 0 255.255.255.240 0

Network switch

A single network switch would be a single point of failure (SPOF). Therefore, a highly available network needs at least two switches with built-in redundancy features. In the sample configuration two CISCO 3750E switches are used. A Link Aggregation group is defined, covering the two switches, which are connected to each other to simulate a single virtual switch. Cisco calls this feature Cross Stack EtherChannel. It supports IEEE 802.3ad and LACP.

AIX network adapter

The AIX environment has redundancy on the VIO Server level and within each VIO itself. There are two VIO servers per physical machine, each with two physical adapters. The adapters are configured as a Shared Ethernet Adapter (SEA) failover setup to achieve physical network adapter redundancy. The secondary adapter takes over when the primary adapter fails.

Advantages and disadvantages

Dedicated high availability tests have shown that this configuration can be an alternative setup to the recommended network setup with OSPF and multiple subnets. Usage of the simplified configuration depends on the customer requirements for network high availability and on the risk the customer can take.

Advantages:

• Simplified network setup with reduced complexity
• No dynamic routing configuration required
• One flat network
• Alternate routes can include HiperSockets. Starting with zEnterprise® 196 GA2 processors, the Internal Queued Direct I/O (IQDIO) support offers another option besides the basic HiperSockets function, which can be used for alternate routes. This functionality is called extended IQD (IQDX). IQDX interfaces can now be dynamically detected. Then, network traffic between systems on the same CEC on a z196 GA2 could use HiperSockets IQDX paths transparently and would use external OSAx paths when necessary.

Disadvantages:

• Redundancy is only possible within the flat network. Other media such as HiperSockets or XCF cannot be used as alternate routes before zEnterprise 196 GA2.
• Redundancy only given in failure cases, which are detectable through hardware. Therefore, redundancy is not achieved, for example, if the OSA Express feature does not detect a loss of signal, TCP/IP has no knowledge of the failure of the network path. In that case it is up to TCP/IP timeouts to detect a failure in the network path and to terminate existing connections. A timeout is not a dedicated and meaningful error to TCP/IP so after a timeout there would be no trigger for TCP/IP to perform the takeover.