6 Dec
2004
6 Dec
'04
2:03 p.m.
Fred N. van Kempen <waltje(a)pdp11.nl> wrote:
I actually wrote a driver that did IP-over-DDMCP on
the VAX end,
Such a driver has existed in standard BSD since 4.2, except that it's
for the DMC-11/DMR-11/DMV-11 (which do all DDCMP processing on board and
can't be made to talk any other protocol to my knowledge) rather than
for the DMF32 that this thread is presumably about.
The DMF32 sync port supports DDCMP too, but it leaves more work to
the host and supports HDLC as well.
with a modified IOS running at the other end, on a
Cisco 2501.
If you needed to connect the DMF32 sync port to a Cisco router, why didn't
you implement Cisco HDLC on the VAX side instead? (As I've done in 2000
in 4.3BSD-Quasijarus0b on a DSV11.) I'm sure you know the frame format:
<flag> <addr> 00 <Ethertype> packet <FCS-16> <flag>
addr is 0F for unicast or 8F for multicast/broadcast*.
It's the simplest and IMO the best framing format for IP over a sync line,
I've used it on T1s (well, fractional T1 since DSV11 is limited to 256 kbps),
and it comes from Cisco.
*The only thing that I never understood is what the hell did Cisco mean
by multicast or broadcast on a point-to-point link. I guess it dates from
the days before people realised that a point-to-point link is not a network
(no ARP, etc.) and does not need to burn up a net/subnet number, and people
would map a net number (hopefully a /30 subnet) to each point-to-point link.
Then .1 and .2 addrs would be assigned to the ends and I guess the 8F addr
byte was generated on packets addressed to .3... My driver always puts
0F in the address field of transmitted packets and treats 0F and 8F as
equivalent on reception; and I always configure the net interface as
unnumbered (for my end use the same address as assigned to one of my real
Ethernets, for the other end put in whatever addr the ISP gave me for
their router; my driver does not require them to be numerically related
as they would be with a /30 subnet).
The approach of using a /30 subnet for a point-to-point link now has to
be used only when you are forced to use an Ethernet as a point-to-point
link without PPPoE, which is stupid but unfortunately used by some DSL
providers.
This is why I usually grab all the sync modems I can
lay my hands on;
USR Courier V.Everything (external of course) supports sync mode on its
DB25 port. The problem is with the other end of the call... I'm using
such modem for my PPP connection, and while I would love to run sync PPP,
I doubt that the ISP's POP will support it.
MS
6 Dec
6 Dec
2:38 p.m.
On Tue, 7 Dec 2004, Michael Sokolov wrote:
Probably because that was a test link; the production system (with the
same driver) connected to a real DDMCP network. The choice was to either
change the WAN to HDLC/PPP, or make that one box talk IP over DDMCP.
Such a driver has existed in standard BSD since 4.2,
except that it's
for the DMC-11/DMR-11/DMV-11 (which do all DDCMP processing on board and
can't be made to talk any other protocol to my knowledge) rather than
for the DMF32 that this thread is presumably about.
Indeed. I run DMV11's
here. I know.
The DMF32 sync port supports DDCMP too, but it leaves
more work to
the host and supports HDLC as well.
Correct. Still, it offloads a LOT of the
enframe and deframe work
to the board CPU.
with a
modified IOS running at the other end, on a Cisco 2501.
If you needed to connect
the DMF32 sync port to a Cisco router, why didn't
you implement Cisco HDLC on the VAX side instead? (As I've done in 2000
in 4.3BSD-Quasijarus0b on a DSV11.) I'm sure you
know the frame format:
<flag> <addr> 00 <Ethertype> packet <FCS-16> <flag>
addr is 0F for unicast or 8F for multicast/broadcast*.
After 17 years of
networking, I guess I'm aware of the various
frame formats, yes.
*The only thing that I never understood is what the
hell did Cisco mean
by multicast or broadcast on a point-to-point link. I guess it dates from
Frame
filtering.
the days before people realised that a point-to-point
link is not a network
(no ARP, etc.) and does not need to burn up a net/subnet number, and people
would map a net number (hopefully a /30 subnet) to each point-to-point link.
Then .1 and .2 addrs would be assigned to the ends and I guess the 8F addr
byte was generated on packets addressed to .3... My driver always puts
There are
very good reasons for assigning IP addresses to both ends of
a WAN connection, especially in multipath environments and/or networks
that use link-state routing protocols. We dont all use "route add
default wan0" you know.
The approach of using a /30 subnet for a
point-to-point link now has to
be used only when you are forced to use an Ethernet as a point-to-point
link without PPPoE, which is stupid but unfortunately used by some DSL
providers.
Actually, by most xDSL providers in Europe. So, apparently it's
not
THAT stupid. Its a cheap 10Mbps (or even 100Mbps, if needed) leased
line technology with baseband encoding to them :)
USR Courier V.Everything (external of course) supports
sync mode on its
DB25 port. The problem is with the other end of the call... I'm using
such modem for my PPP connection, and while I would love to run sync PPP,
I doubt that the ISP's POP will support it.
If they use TotalControl or akin
racks, they will. I used to run that
over my USR Courier I-Modem (ISDN V.Everything with both digital and
analog protocols.)
--f
2:56 p.m.
On Tue, 7 Dec 04 01:03:17 GMT, Michael Sokolov <msokolov(a)ivan.harhan.org> wrote:
If you needed to connect the DMF32 sync port to a
Cisco router, why didn't
you implement Cisco HDLC on the VAX side instead? (As I've done in 2000
in 4.3BSD-Quasijarus0b on a DSV11.) I'm sure you know the frame format:
<flag> <addr> 00 <Ethertype> packet <FCS-16> <flag>
addr is 0F for unicast or 8F for multicast/broadcast*.
It's the simplest and IMO the best framing format for IP over a sync line,
I've used it on T1s (well, fractional T1 since DSV11 is limited to 256 kbps),
and it comes from Cisco.
Hmm... an interesting idea... I have these COMBOARDs, see... and they
have a 68000 processor on board, and either a COM5025 or a Z8530
serial chip. I seriously doubt that the processor could keep the line
fed much faster than 128kbps or 256kbps (99% of our customers needed
9600 - 56/64k), at least it has a DMA interface on the VAX side.
The boards have between 128K and 2MB of DRAM, plus 16K - 128K of ROM
(and are designed to get their application into RAM from a file on the
VAX)... I think there's
everything there except the app to make them into HDLC IP conduits...
I might have to fiddle with that a bit when I get back. I'd been
thinking of a way to make these guys talk IDE or SCSI... to at least
have _some_ practical use for them.
-ethan
7 Dec
7 Dec
9:03 a.m.
*The only thing that I never understood is what the
hell did
Cisco mean
by multicast or broadcast on a point-to-point link. I guess
it dates from
the days before people realised that a point-to-point link is
not a network
Well there certainly were non-point-to-point synch implementations
around: DEC DDCMP supports multi-drop links.
(no ARP, etc.) and does not need to burn up a
net/subnet
number, and people
would map a net number (hopefully a /30 subnet) to each
point-to-point link.
There's an RFC around somewhere that allows for /31 subnets
for such purposes (.1 and .0 being allowed as the two ends).
No idea if it ever caught on though.
Antonio
--
---------------
Antonio Carlini arcarlini(a)iee.org
10:02 a.m.
dates from the
days before people realised that a
point-to-point link
is not a network
Well there certainly were non-point-to-point synch implementations
around: DEC DDCMP supports multi-drop links. There's an RFC around somewhere that allows for
/31 subnets
for such purposes (.1 and .0 being allowed as the two ends).
No idea if it ever caught on though.
Eep, I'm just getting used to /30, don't complicate things :)
cheers
w
8 Dec
8 Dec
12:57 p.m.
On Tue, 7 Dec 2004 20:03:54 -0000, Antonio Carlini
<a.carlini(a)ntlworld.com> wrote:
Well there certainly were non-point-to-point synch
implementations
around: DEC DDCMP supports multi-drop links.
I think exactly once, one of our customers put a COMBOARD into a
multi-drop environment... worked fine, but the salesguy was nervous
since it would work theoretically, but we hadn't done exhaustive
testing.
-ethan
7 Dec
7 Dec
9:09 a.m.
It's the simplest and IMO the best framing format
for IP over
a sync line,
I've used it on T1s (well, fractional T1 since DSV11 is
limited to 256 kbps),
and it comes from Cisco.
Nearly missed this bit. The DSV-11 was rated at 256kbps but
it can actually do either a tiny smidgen over (or a tiny
smidgen under) 1Mbps on a worst case HDLC sequence,
essentially forever.
There was no supported service around that speed (I think
the UK Megastream was 1.024Mbps) so it was sold as
256kbps.
Feel free to overclock it - it should handle it
just fine!
Antonio
--
---------------
Antonio Carlini arcarlini(a)iee.org
7161
days inactive
7163
days old
6 comments
5 participants
participants (5)
-
a.carlini＠ntlworld.com -
ethan.dicks＠gmail.com -
msokolov＠ivan.Harhan.ORG -
waltje＠pdp11.nl -
witchy＠binarydinosaurs.co.uk