Alan and Mark, Thanks for the help. I haven't tried it yet, but I suspect that if I change the ip addresses on the new hipersocket interfaces, it will help.
I think part of my confusion came from reading section 4.3 of the Linux for IBM System z9 and IBM zSeries redbook and mis-interpreting the paragraph regarding the hipersockets not using LAN frame, instead they are addressed by data queue addresses. I guess I read too much into it and that I took that to say that the mac address and ip address are no longer part of that data stream and that the ip stack knew that destination by an internal device address instead of ip address and did not use normal ip routing. Did I read it incorrectly and is the network layer info (dest ip address etc) still in the frame of data? So that a network concentrator like Linux on z/OS just needs to change the frame protocol to go out the appropriate interface much a like a router supporting token ring and Ethernet. It looks like I need to give it more thought to determine just how many Linux guest will want hipersocket access to z/os data. It may not be worth the effort to configure a hipersocket guest lan for the the Linux guests to connect into, just save on real hipersocket addresses. If you are curious to see a simple network map of what we were considering, I can send it offline. Thanks, Ken Spracklen -----Original Message----- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Alan Altmark Sent: Thursday, November 15, 2007 09:12 To: LINUX-390@VM.MARIST.EDU Subject: Re: Hipersocket Guest Lan Question On Wednesday, 11/14/2007 at 06:00 EST, "Spracklen, Ken" <[EMAIL PROTECTED]> wrote: > Recently we started planning for Linux guests to access data on the > z/os lpars via hipersockets. As part of that process, we defined a new > hipersocket, IUTIQDEF, to the z/VM TCPIP guest profile and added the > IUTIQDEF device to the OSPF config for MPROUTE (same subnet as the > IUTIQDFF interface). We then created a new guest lan (glan01 with type > hipersocket and used nicdef/couple commands for a Linux guest and the > TCPIP guest to use that hipersocket guest lan. > > The problem is that I could not ping the ip address of z/VM IUTIQDEF > interface from the zOS systems nor ping the hsi1 ip address on the > Linux guest. A q lan glan01 det indicated both guests connected but > the TX/RX counts were 0 and the discards were high. If you draw the picture Rob requested, you will see that a Guest LAN is a separate subnet and needs to be attached to your existing network using the same techniques as you would for a real LAN segment. > 1) Is the above scenario possible or recommended? We are hoping to use > the guest lan concept to conserve on real hipersocket devices needed > for Linux guests. Does routing work the same way (via the a lookup > table) for a hipersocket guest lan as it does on a real hipersocket network? It is fine, though you may wish to give Linux OSA Guest LAN (QDIO or Hipersocket) is not a bridge or switch. It is a LAN segment that requires routing. > 2) Would the above scenario work if the new guest lan was configured > as QDIO? Is there much savings in cp cycles and memory by using a > hipersocket guest lan vice a qdio guest lan? That depends on the workload. The z/VM Performance Reports, going back to z/VM 4.3, provide a comparison of QDIO vs. HiperSockets. > 3) In the documentation we have encountered, we have seen examples of > a z/OS hipersocket concentrator and a Linux network concentrator. Does > someone have an example of a z/VM TCPIP/MPROUTE concentrator? Can z/VM > TCPIP/MPROUTE support routing between hipersockets, osa's, and guest > lan (hipersockets and/or qdio)? Which type of concentrator is recommended? z/VM does not provide a HiperSocket concentrator (i.e. a HiperSocket Virtual Switch). > 4) In the scenario of the z/os hipersocket concentrator, the z/os is > the router and there are a few Linux guests attached to the > hipersockets. I was wondering if someone could clear up my > understanding how that Linux guests can route data to an ip address > out on the network beyond z/os tcpip. In my foggy mind, I can see the > Linux guest indicating the default route is out the hipersocket > interface, but if the hipersocket frame doesn't have the LLC info like > the ip address of the destination, how does z/os know what ip address > to put into the destination ip address so that goes out into the > "normal" lan network? Or is my understanding of hipersockets incorrect > about the LLC headers It's not a "routing" thing. Linux thinks the same-subnet IP addresses on the OSA network are local and does what Linux normally does: sends the packet directly to the intended IP address, not via a default route. A miracle occurs in Step 2 and z/OS gets the packet, unchanged, and places it on the LAN. In return, z/OS has responsibility to handle ARP issues on the OSA on Linux's behalf. To the rest of the network, z/OS and all of the Linuxen have the same MAC address. Alan Altmark z/VM Development IBM Endicott ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
