Allan Thanks for covering one of the important points about arranging to be able to use multiple paths for the *inbound* traffic.
The destination IP address used by the partner IP nodes, perhaps "clients" where Mark's z/OS system is the "server", is going to need *not* intimately to be associated with any of the possible real interfaces but is going to need to be a *virtual* interface - perhaps one for each of the "server" applications, "services", supported by the z/OS system if proper planning for existing or future flexibility is being considered. Only where the OSA feature port interface IP addresses are, or appear to be, "en route" to the destination IP address can some common router on the path between the partner node, the "clients", and the "server" node spread the traffic over equivalent paths, always with the requirement that they are assigned equal "cost". Chris Mason On Wed, 9 Jun 2010 12:17:11 -0500, Staller, Allan <[email protected]> wrote: >Would a VIPA provide the functional equivalent of "link aggregation"? > >I am referring to the process of gaining greater aggregate thruput via >the use of >one or more physical interfaces, as opposed to a RFC like definition of >"link aggregation". > >Something like the following (assuming the same VLAN for both adapters): > > > IPCONFIG MULTIPATH PERPACKET (PERCONNECTION?) > vipa definitions go here > > DEVICE xxxxxxx MPCIPA NONROUTER > LINK ETHERyyyy IPAQENET xxxxxxx > > DEVICE aaaaaaa MPCIPA NONROUTER > LINK ETHERbbbb IPAQENET aaaaaaa > > >HOME > 192.168.81.41 ETHERyyyy ;static > 192.168.81.235 ETHERbbbb ;static > ;192.168.81.40 vipa > ><snip> >... If you have multiple OSA feature ports connected to the network in >such >a way that the same destination IP address can be reached by more than >one of >your routing table entries, by means of the IPCONFIG statement MULTIPATH >parameter >you can cause outgoing traffic to use more than one OSA feature port to >"feed" the >same destination. There is still the choice between the PERPACKET and >PERCONNECTION subparameter. Clearly PERPACKET achieves a more even >distribution but there are some "philosophers" who prefer PERCONNECION >for >reasons I forget right now.[3] > > [3] > >http://publibz.boulder.ibm.com/cgi- bin/bookmgr_OS390/BOOKS/F1A1B491/2.38 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
