David Boyes wrote:
>
> 2) During migration to the best way; Can my current LINUX
> image share the
> OSA-E with zVM?
>        And, if yes, how? I can't seem to find an example of
> sharing a OSA-E
> in the books.
>      I have seen the warnings about the need for both OS to specify
> portname. But,
>       for example, if LINUX is using address x00,x01,x02;
> does zVM need to
> specify
>       a different address on the DEVICE statement.

Sharing OSA-E cards is quite simple as the card itself sets up the sharing.
No external software, such as OSA/SF is needed. The complications come
mostly from OSA2 cards, which require external an external OSA/SF tool to
set up the sharing.

1) Gen the OSA-E card as OSD for QDIO (GBX or FEX) or OSE for LCS  (FEX
only) (I'm using QDIO with the latest 2.2.16 modules) and indicate that
they are SHARED between LPARS.

2) If you only use one TCP/IP IP address per LPAR, then you can gen for 4
subchannel addresses for QDIO. Each LPAR will see the same 4 addresses (for
example: 4D00, 4D01, 4D02, 4D03 each LPAR) and can code the same addresses
as a "row" in the OSA card OAT table contain LPAR name, subchannel address,
and IP.

      example: this is automatically set up by the OS-E card as the LPAR
asks for the device
      LPAR        subchannel address      IP address
      lpar1       4d00              10.32.86.91
                  4d01              10.32.86.91
                  4d02              10.32.86.91

      lpar2       4d00              10.32.86.92
                  4d01              10.32.86.92
                  4d02              10.32.86.92

3) The only real problem occurs when you need more that 1 TCP/IP IP address
per LPAR because the addresses are assigned at the CHP level so that each
LPAR sees the same number of subchannel addresses. The formula of maximum
subchannel addresses per LPAR is

      There is a maximum of 240 subchannel addresses on an OS-E card which
each LPAR seeing the same number of addresses.

      subchannel addresses/LPAR = (240/number of LPARS)

The IOCDS should then reflect the result. Example: 2 LPARS could be genned
with 120 subchannels each or  15 LPARS could be genned with 16 subchannel
addresses.

The the number of subchannel addresses needs to be devided up between IP -
2 subchannel addresses for LCS per IP, 3 subchannel addresses for QDIO
under VM per IP, and 4 subchannel addresses for QDIO under VM per IP.

Assuming 1 Linux and 1 VM LPAR, the address assignments may be something
like:

      LPAR        subchannel address      IP address
      lparx       4d00              10.32.86.91
                  4d01              10.32.86.91
                  4d02              10.32.86.91

      lparvm            4d00              10.32.86.92 vm device   900
                  4d01              10.32.86.92             901
                  4d02              10.32.86.92             902
                  4d03              10.32.86.93             904 < linux
must start on even!
                  4d04              10.32.86.93             905
                  4d05              10.32.86.93             906

And no, under this scheme, the LPAR that needs the most subchannel/IP
addresses cannot use the addresses from the other LPARS, even if they are
not used!

As to portname, that is LIC dependent and affects all OS. There is a WCS
flash that talks to it. The most often recurring problem seems to be that
the CASE has to be the same on all OS and the first OS to IPL sets the
portname. So, if VM uses upper case for the portname, Linux must too!

Regards, Jim
Linux S/390-zSeries Support, SEEL, IBM Silicon Valley Labs
t/l 543-4021, 408-463-4021, [EMAIL PROTECTED]
** Grace Happens **

Reply via email to