T.R | Title | User | Personal Name | Date | Lines |
---|
928.1 | DEFPA? | SWAM1::POIANI_MI | | Wed Apr 30 1997 11:25 | 2 |
| The SPD onRIS Bootp says it supports the DEFPA FDDI cards, are the
ATM cards not DEFPA cards as well?
|
928.2 | | WASNT::"[email protected]" | Born to be Mild | Wed Apr 30 1997 12:37 | 3 |
| The DEFPA cards are FDDI cards, not ATM cards.
doug
|
928.3 | RIS over ATM...... Sorry, "NO GO" lights | KAMPUS::BRANDNER | That's noot eentiiirly aaacuuraate ... | Wed Apr 30 1997 12:55 | 38 |
|
How does RIS work?
If you do something like
>>> boot -protocols bootp ewa0
the system broadcasts a bootp request....
^^^^^^^^^^
Now to ATM. ATM is a Non Broadcast Multiple Access (NBMA) media.
You would need some prerequisites to do a broadcast over ATM
(you basically have to emulate multicasts and broadcasts):
You need a Multicast Address Resolution Server (MARS), or at least
a Multicast Server (MCS), the NSAP of that MARS/MCS, an up and
running ATM subsystem with signalling (ILMI) enabled, a VC to the
MARS/MCS.
Then you could send a bootp request to the MARS/MCS which in turn would
send it to all registered end systems. If your
But I don't think that all the prerequisites (MARS, running ATM
subsystem, signalling, etc.) would nicely fit into the console code
(which is already about 2 MB, when memory serves right). So this
is no UNIX issue for the RIS client.
Not to mention that there's no MARS implementation for
DIGITAL UNIX <whatever_version> currently available.
Nevertheless, it would be definately breath taking to do a RIS install
over ATM regarding the speed!
But NFS over ATM runs just fine (up to the point when you lose the
ATM connection to the NFS client/server. NFS can't imagine that it
cannot write to a device which is the case when you lose the VC or PPA.
Even if your VC to the NFS clien/server comes back online, NFS is still
puzzeled about a failure of the write call and hangs there).
Rudi
|
928.4 | | SMURF::PSH | Per Hamnqvist, UNIX/ATM | Wed Apr 30 1997 14:16 | 9 |
| | the system broadcasts a bootp request....
| ^^^^^^^^^^
While this is not supported with our gear, you do not actually
have to broadcast. I have seen ATM based set top boxes that have
their "bootp" server NSAP in PROM/Flash and then just sets up an SVC
directly to it to load.
>Per
|
928.5 | Client or server on ATM? | SMURF::GREBUS | DTN 381-1426 - Digital Unix | Wed Apr 30 1997 14:59 | 10 |
| We don't support RIS installing or running diskless on systems with
only an ATM adapter. There is no console support for booting over ATM,
and as Rudi mentioned it would be a significant task just to fit it in.
If the clients are on Ethernet, FDDI, etc and it is the server that has
only ATM, then it should work, provided the router in between can act
as a "bootp relay agent".
/gary
|
928.6 | | NPSS::NEWTON | Thomas Newton | Wed Apr 30 1997 17:13 | 9 |
|
Re: .4
If you wanted to support SVC-based booting, you wouldn't even need to
configure the server address in every workstation.
Just configure it once in each switch, and have the workstations read
it from the ILMI Service Registry MIB, like LECs do. Benefit: if the
server address changes, you'd have many fewer copies to update.
|
928.7 | | SMURF::PSH | Per Hamnqvist, UNIX/ATM | Wed Apr 30 1997 20:17 | 15 |
| | Just configure it once in each switch, and have the workstations read
| it from the ILMI Service Registry MIB, like LECs do. Benefit: if the
| server address changes, you'd have many fewer copies to update.
That is a good idea. The environment I was talking about did not have ILMI.
Also, the slight drawback with using ILMI as the only source is that offers
only one boot server for all systems connected to that switch (unless you
have an elaborate negotation protocol) and also assumes that the first switch
will actually know; your first switch may be nothing but a front end, like
a 6 port Virata. Perhaps a combination of the two would work: check prom
first and if nothing, default to ILMI?
Cheers,
>Per
|
928.8 | | NPSS::NEWTON | Thomas Newton | Wed Apr 30 1997 21:09 | 14 |
|
> Also, the slight drawback with using ILMI as the only source is that offers
> only one boot server for all systems connected to that switch (unless you
> have an elaborate negotiation protocol)
That depends upon the implementation. Ours will let you set up things so that
one group of workstations sees addresses A and B, while another sees address C.
Each peer ATM node thinks it sees "THE" Service Registry MIB, when in fact, it
sees the view that the switch wants it to see.
> Perhaps a combination of the two would work: check prom first and if
> nothing, default to ILMI?
Yes.
|
928.9 | | NPSS::NEWTON | Thomas Newton | Wed Apr 30 1997 21:18 | 6 |
|
For the case where the front-end switch does not implement the Service Registry
MIB, you could use a well-known ATM address.
You'd probably have to submit a SVC-boot proposal to the ATM Forum to get one -
but having a standardized version might not be all that bad.
|
928.10 | Will we ever support it? | SWAM1::POIANI_MI | | Thu May 01 1997 23:08 | 7 |
| Will we be supporting RIS/bootp/diskless in the future? It seems it's been
identified how to do it, is this something we will be doing?
SUN would not commit to my customer (Sandia Labs) that they would do
it.
Mike p.
|
928.11 | No plans to support it | SMURF::GREBUS | DTN 381-1426 - Digital Unix | Fri May 02 1997 14:41 | 19 |
| Some aspects of "how to do it" have been identified. There are more,
in particular, how do you cram an ATM driver, signalling stack,
ILMI and RFC1577 into the 2 MB of memory available to the console
(which is already heavily overlayed and has trouble fitting on large
configurations).
Also, it's unlikely to be supported across all platforms...there aren't
console development resources available to make such major changes to
older consoles. Actually, I don't believe all platforms currently
support diskless operation, although I think they all support ris.
I'm not aware of any current plans to support or to investigate supporting
RIS/bootp/diskless over ATM. Since this would involve significant
development work for the console (hardware) groups, I suspect it will
only happen if someone makes the case that it will generate significant
incremental hardware revenue.
/gary
|