I appreciate the elaboration and its relevance to link aggregation.  My 

next question would be how to move toward non-shared OSAs and maintain no
n-
VLAN-aware guests connected to multiple VLANs through the OSA.  The two 

seem to be incompatible.  Either the guests need to become VLAN-aware to 

have multiple NICs to the same VSWITCH, or NICDEF needs a new parameter t
o 
define its' default VLAN (to supercede the VSWITCH GRANT default VLAN for
 
*every* NIC from a guest).

Is there another way?

Brian Nielsen

On Mon, 26 Mar 2007 12:16:00 -0400, Alan Altmark <[EMAIL PROTECTED]
> 
wrote:

>On Monday, 03/26/2007 at 09:53 EST, Brian Nielsen <[EMAIL PROTECTED]
V>
>wrote:
>> On Fri, 23 Mar 2007 09:55:05 -0400, Alan Altmark
><[EMAIL PROTECTED]>
>> wrote:
>>
>> >It is also my
>> >recommendation that you do not share an OSA in use by a VSWITCH,
>> >*especially* if it is VLAN-aware.  The last thing you want is a fight

>> >between two VSWITCHes about whether a VLAN is in use.  Yes it is, no 
it
>> >isn't, yes it is, no it isn't....  If you choose to go down this path
,
>> >define the VSWITCH with NOGVRP.
>>
>> Would you please eloborate on this?  I have multiple VLAN-aware
>VSWITCHes
>> sharing the same OSA with no problems.  Each VSWITCH has a distinct an
d
>> separate set of VLANs, if that makes any difference.  This setup is
>> required to connect non-VLAN-aware guests to multiple VLANs via VSWITC
H
>> GRANTs.
>>
>> If there is a pitfall in there somewhere I'd like to know more about i
t.
>
>As long as you do not try to span VLANs across different VSWITCHes shari
ng
>an OSA, it will work.  The problem I run into is that someone changes th
e
>switch (virtual or real) or VLAN configuration without thinking about th
e
>affect on a shared port.  Things start to fail.  But it's just a
>recommendation, not a requirement.  Our gun.  Your foot.
>
>In z/VM 5.3 we announced support for IBM System z9 IEEE 802.1ad Link
>Aggregation ("channel bonding", "etherchannel"), providing additional
>bandwidth with each OSA you add to the "aggregation group".  In this
>configuration, the OSAs are usable by only a single VSWITCH at a time, a
nd
>will use additional hardware support to enforce this.  (There is a
>switch-to-switch protocol flowing over the links along with user data an
d
>more than one cook in the kitchen will spoil the broth...)   Imagine up 
to
>80 Gb of data per second moving in/out of the box (up to eight 10 Gb
>ports).
>
>This support also provides for the smooth addition or removal of OSAs fr
om
>the aggregation group, whether because you want to add/take an OSA, or
>because one fails.  No data is lost, with retransmission handled at a
>packet level rather than at the TCP segment level.
>
>If you ever think you will want to take advantage of this support, you
>need to be planning for dedicated OSAs, even if you need to share right
>now.
>
>And, of course, mixing VLAN-aware and non-VLAN-aware is a no-no, though
>nothing will enforce it.
>
>So, none of this is really about what *can* be done.  You can share all
>you want.  But the risk to your virtual data center is lowest if you don
't
>share, and that remains my recommendation.
>
>Admittedly, "because Alan says so" only goes so far (the "Wesayso
>Corporation", anyone?).  I have the same problem at home.  Go figure.  :
-)
>
>Alan Altmark
>z/VM Development
>IBM Endicott
>========================
=========================
========================

Reply via email to