Remove VAXes from Cluster and FDDI

[Martin Hughes]

As I understand it, SCS traffic will simply go via the fastest path available. The path will be chosen in the order listed below.

1. Memory Channel (MCDRIVER)
2. CI (PADRIVER)
3. DSSI (PADRIVER)
4. FDDI/Ethernet (PEDRIVER)

In the case of 4. PEDRIVER will compare latency and choose the fastest path.

[Art Wiens]

Well thanks everyone, it was a good learning day for me! I read the Cluster guidelines manual and realize that I've missed the "sublety" of the difference between the types of interconnects -

Cluster interconnects (node to node only)
Ethernet (10/100, Gigabit)
Asynchronous transfer mode (ATM)
FDDI (Fiber Distributed Data Interface)
MEMORY CHANNEL
Shared Memory CI (SMCI) (Galaxy instance to Galaxy instance)

Shared-storage interconnects (node to storage only)
Fibre Channel
SCSI (Small Computer Systems Interface)

Both node-to-node and node-to-storage interconnects
CI (computer interconnect)
DSSI (Digital Storage Systems Interconnect)

And that both ethernet and fddi use the PEDRIVER ... SDA is showing me a virtual circuit.

So now with my new knowledge of the hw config having MC, and clearer picture of how SCS can work, perhaps it's ok to yank fddi altogether. The nic and fddi are both 100Mb and it theory says it will use MC anyways. Ethernet can remain as a redundant SCS path but there is no need to MSCP serve the HSx disks to "anyone" at 100Mb any more.

Thanks again,
Art

[Anton van Ruitenbeek]

Art,

To add a small issue to it:
FDDI is already done by a concentrator, so you have only to remove the two VAXes. As Steve already mentioned, FDDI is a dual system. You only will see messages in OPCOM that you remove next/previous node and automaticly will see another next/previous node.
FDDI and NIC are theoreticly the same speed. Because of the nature of FDDI, this will be effective about 80Mb and NIC's isn't realy to tell. This is depending on the brance of routers you have and ofcourse the traffic. I know a big problem Cisco has, is that when you have a flame of traffic, Cisco drops _ALL_ other traffic than IP. This is crusial for clusters because they need SCS. You also using MC and this is the prefered path because its the fastest. Only also the shortest expand.
Whe didn't use MC because its to expensive for the need and the distance is to short. We have a cluster over more then 7 km. Thats wy we have FDDI (and Digital concentrators and fallback to 100Mb NICs (with unfortunaly Cisco concentrators).

AvR