25 Jan
2016
25 Jan
'16
4:25 p.m.
After having my links up and running a couple of days now (and they work
fine), I've come to realize that if we want to have a somewhat efficient
topology, we need to agree on circuit costs.
Now, that said, in todays internet world, network speeds are fast enough
that this isn't really a big problem. So if people don't want to, I'm
okay with that.
But otherwise we should try and use some agreed upon scheme for setting
costs on circuits. The best suggestion I've seen so far is to base it on
ping times. Would people be willing to adopt this and implement it all
over the place for level 2 routers? Because, if not everyone agrees,
then I'm wasting my time here. :-)
The exact cost scale we can discuss, if we agree on doing it. But it
needs to be pretty much used by everyone, or else it's just a waste of
time and energy.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
25 Jan
25 Jan
4:59 p.m.
On Jan 25, 2016, at 10:25 AM, Johnny Billquist <bqt
at softjar.se> wrote:
After having my links up and running a couple of days now (and they work fine), I've
come to realize that if we want to have a somewhat efficient topology, we need to agree on
circuit costs.
Now, that said, in todays internet world, network speeds are fast enough that this
isn't really a big problem. So if people don't want to, I'm okay with that.
But otherwise we should try and use some agreed upon scheme for setting costs on
circuits. The best suggestion I've seen so far is to base it on ping times. Would
people be willing to adopt this and implement it all over the place for level 2 routers?
Because, if not everyone agrees, then I'm wasting my time here. :-)
Sounds good to me.
Given that things are basically fast, just setting all costs to the same number would not
be all that horrible, either. But I agree with you that in today's networks the only
really interesting metric is latency.
paul
5:09 p.m.
On 2016-01-25 16:59, Paul_Koning at Dell.com wrote:
Yes, my other options that I am considering is to just set all costs
equal on my local system. And if we don't get a unified cost metric
system implemented, that's what I'm going to do.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
On Jan 25, 2016, at 10:25 AM, Johnny Billquist
<bqt at softjar.se> wrote:
After having my links up and running a couple of days now (and they work fine), I've
come to realize that if we want to have a somewhat efficient topology, we need to agree on
circuit costs.
Now, that said, in todays internet world, network speeds are fast enough that this
isn't really a big problem. So if people don't want to, I'm okay with that.
But otherwise we should try and use some agreed upon scheme for setting costs on
circuits. The best suggestion I've seen so far is to base it on ping times. Would
people be willing to adopt this and implement it all over the place for level 2 routers?
Because, if not everyone agrees, then I'm wasting my time here. :-)
Sounds good to me.
Given that things are basically fast, just setting all costs to the same number would not
be all that horrible, either. But I agree with you that in today's networks the only
really interesting metric is latency. 
6:08 p.m.
Paul_Koning (Paul_Koning at Dell.com) wrote:
just setting all costs to the same number would not be all that horrible,
either
The effect of this would be to make DECnet prefer the route with the least
number of hops. That makes a lot of sense to me and I would absolutely
agree, BUT there's a problem...
The problem is the HECnet bridge program. The bridge effectively makes
everything adjacent to everything else, even though the actual communication
goes thru psilo. Thus things look to DECnet like they're only one hop away
when in reality they're at least two.
The consequence of this is that two machines which both have a connection
via the bridge and which have a direct point-to-point link between them
(e.g. Multinet) will sometimes decide to talk via the bridge and will ignore
the direct link. At least for my part, that not the behavior I want.
Bob
6:13 p.m.
On 2016-01-25 18:08, Robert Armstrong wrote:
Yes, this would in theory end up using the least number of hops, but
that would depend on everyone else also setting the same equal cost on
all their circuits, otherwise the bets are still off.
Paul_Koning
(Paul_Koning at Dell.com) wrote:
just setting all costs to the same number would not be all that horrible,
either
The effect of this would be to make DECnet prefer the route with the least
number of hops. That makes a lot of sense to me and I would absolutely
agree, BUT there's a problem... The problem is the HECnet bridge program. The
bridge effectively makes
everything adjacent to everything else, even though the actual communication
goes thru psilo. Thus things look to DECnet like they're only one hop away
when in reality they're at least two.
The consequence of this is that two machines which both have a connection
via the bridge and which have a direct point-to-point link between them
(e.g. Multinet) will sometimes decide to talk via the bridge and will ignore
the direct link. At least for my part, that not the behavior I want.
On the other hand, the traffic passes through a lot less inside your VMS
box... Most of the network processing at psilo is done by hardware, and
depending on what you are using at your local end to run the bridge is
the same thing, so in fact it might very well be faster than the
Multinet tunnel...
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
6:28 p.m.
Most of the network processing at psilo is done by
hardware,
... or, just make the cost on the Ethernet circuits higher than the cost on the direct
point to point circuits. This is the complete opposite of the traditional situation, but
the bridge isn't a traditional LAN.
FWIW, I think this is all Peter wanted when he brought up the topic weeks ago.
We've come full circle now.
Bob
6:35 p.m.
On 2016-01-25 18:28, Robert Armstrong wrote:
The problem is that the goals that I've gotten the impression you and
Peter have are different to what I want to achieve.
And I feel that we're not getting anywhere, so I'm discarding the idea.
I'll just set my costs all equal, and it will work well enough.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
Most of the
network processing at psilo is done by hardware,
... or, just make the cost on the Ethernet circuits higher than the cost on the
direct point to point circuits. This is the complete opposite of the traditional
situation, but the bridge isn't a traditional LAN.
FWIW, I think this is all Peter wanted when he brought up the topic weeks ago.
We've come full circle now.
6:43 p.m.
the goals that I've gotten the impression you and
Peter have are
different to what I want to achieve.
Actually Peter just talked to me at lunch about his problem last month,
but we're not working together. He has some internal network issue that I
won't pretend to understand. Peter can speak up about it if he wants.
My goals are just what I said - only two, and I was even willing to drop
one.
So you feel it's impossible to set up costs so that a direct link is
preferred over the bridge?
Bob
7:31 p.m.
On 2016-01-25 18:43, Robert Armstrong wrote:
Fair enough.
the goals that
I've gotten the impression you and Peter have are
different to what I want to achieve.
Actually Peter just talked to me at lunch about his problem last month,
but we're not working together. He has some internal network issue that I
won't pretend to understand. Peter can speak up about it if he wants. My goals are just what I said - only two, and I was
even willing to drop
one.
So you feel it's impossible to set up costs so that a direct link is
preferred over the bridge?
I feel that it is impossible to setup some some systematic approach to
costs when one sort of direct link is considered a general problem.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
7:40 p.m.
I feel that it is impossible to setup some some
systematic approach to
costs when one sort of direct link is considered a general problem.
I don't understand what that means. Aren't all direct links the same, at least
as far as DECnet cost is concerned?
Unless you're saying that you consider the bridge to be a "direct link"
despite its extra hop thru psilo. If we go that way then any and all links are equivalent
and you've rendered my only goal superfluous.
Bob
7:47 p.m.
On 2016-01-25 19:40, Robert Armstrong wrote:
And I'm not sure I understand that sentence. :-)
All links are the same, I guess. They all have a cost associated. The
cost might be different. Which would suggest they are not the same. So
what do you mean?
I feel that it
is impossible to setup some some systematic approach to
costs when one sort of direct link is considered a general problem.
I don't understand what that means. Aren't all direct links the same, at
least as far as DECnet cost is concerned? Unless you're saying that you consider the
bridge to be a "direct link" despite its extra hop thru psilo. If we go that
way then any and all links are equivalent and you've rendered my only goal
superfluous.
Yes, I consider the bridge to be a direct link as well. All our traffic
is carried over IP, meaning it is running through a number of hops in
between that are invisible. The host running the bridge is not, in my
view, any different than any router, or the extra layers that a Multinet
link is running through inside the VMS box, in order to carry over the
DECnet traffic into an encapsulated UDP or TCP stream.
If the encapsulation happens inside the machine, or externally, how much
do that matter? Especially when the external encapsulation might in fact
be way faster?
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
7:56 p.m.
The host running the bridge is not, in my view, any
different than any router,
Yeah, but my router is in the garage and I can configure it as I need. Psilo is a
router that I have no influence over PLUS it's on the other side of the planet. Those
are not the same in my mind.
But to be fair, I don't even use the bridge and now that you have TCP/IP tunneling
for RSX I can't imagine that I'd ever need to. So as the saying goes, "I
don't have a horse in this race." I drop all my requirements.
Bob
8:01 p.m.
On 2016-01-25 19:56, Robert Armstrong wrote:
In all fairness, you have no control over most of the internet. :-)
Including just about all the hops when you are using a Multinet tunnel
as well.
The host
running the bridge is not, in my view, any different than any router,
Yeah, but my router is in the garage and I can configure it as I need. Psilo is a
router that I have no influence over PLUS it's on the other side of the planet. Those
are not the same in my mind. But to be fair, I don't even use the bridge and
now that you have TCP/IP tunneling for RSX I can't imagine that I'd ever need to.
So as the saying goes, "I don't have a horse in this race." I drop all my
requirements.
A valid point. That said, I've noticed that the link to LEHATO isn't
exactly what I'd call stable. Want to move over to TCP? I suspect it's a
simple case of the occasional packet being dropped, and Multinet links
over UDP just do not deal will with that situation. Meaning, at the
moment, the link between MIM and LEGATO is probably a bit sucky, and if
someone decided to try transferring a larger file over it, chances are
that it will fail repeatedly.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
6:31 p.m.
On Jan 25, 2016, at 12:08 PM, Robert Armstrong <bob
at jfcl.com> wrote:
That's true. So set a bridge circuit cost to 2N if everyone else is N, or if using
delay-based costs, set it to 2 times the average delay based cost between bridge and its
clients.
As more datalink support is being added to various places, it might be worth while moving
away from bridge connections where that is reasonable. Multinet over TCP is a good
alternative. GRE is ok, too. DDCMP (as, for example, implemented in SIMH V4.0) is good.
paul
Paul_Koning (Paul_Koning at Dell.com) wrote:
just setting all costs to the same number would not be all that horrible,
either
The effect of this would be to make DECnet prefer the route with the least
number of hops. That makes a lot of sense to me and I would absolutely
agree, BUT there's a problem...
The problem is the HECnet bridge program. The bridge effectively makes
everything adjacent to everything else, even though the actual communication
goes thru psilo. Thus things look to DECnet like they're only one hop away
when in reality they're at least two.
6:50 p.m.
So set a bridge circuit cost to 2N if everyone else
is N,
I'm completely in favor of that too, but I don't know if it will fly. The
issue is that it's impossible to separate the "bridge circuit" cost from
any
other Ethernet cost, so the bridge cost applies to any other DECnet machines
on your LAN. In almost all cases this is of no consequence, but ISTR there
were some people that this caused a problem. If I'm wrong, then great -
problem solved...
Bob
6:07 p.m.
On 01/25/2016 10:25 AM, Johnny Billquist wrote:
After having my links up and running a couple of days
now (and they work
fine), I've come to realize that if we want to have a somewhat efficient
topology, we need to agree on circuit costs.
Now, that said, in todays internet world, network speeds are fast enough
that this isn't really a big problem. So if people don't want to, I'm
okay with that.
But otherwise we should try and use some agreed upon scheme for setting
costs on circuits. The best suggestion I've seen so far is to base it on
ping times. Would people be willing to adopt this and implement it all
over the place for level 2 routers? Because, if not everyone agrees,
then I'm wasting my time here. :-)
The exact cost scale we can discuss, if we agree on doing it. But it
needs to be pretty much used by everyone, or else it's just a waste of
time and energy.
I will participate.
-Dave
--
Dave McGuire, AK4HZ
New Kensington, PA
6:17 p.m.
I will participate.
Oh, and sure - I'll go along provided I can get the routing results that I want.
Sorry to be selfish, but I'm not going to put up with some nonsensical routing path
just so that we can have standardized costs.
I already put forth my requirements for the routing a while ago - there are only two.
a) If two nodes have a direct point to point link, then that link should be the
preferred route.
b) I really want symmetrical routing.
I already know that there are those who believe that symmetrical routing is impossible to
achieve, so maybe I'll give up on (b). I still want (a) though.
Bob
6:33 p.m.
On 2016-01-25 18:17, Robert Armstrong wrote:
Symmetrical routing will be impossible. DECnet is not designed in a way
that makes it possible.
I can accept a, but I'm not convinced it makes life better, but more to
the point - if you want starting to tweak things based on different,
unrelated things, this will not work, and we might as well give up right
away.
I'm starting to feel that I'm just going to drop this. Do whatever you
want. I'll put all my circuits at the same cost.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
I will
participate.
Oh, and sure - I'll go along provided I can get the routing results that I want.
Sorry to be selfish, but I'm not going to put up with some nonsensical routing path
just so that we can have standardized costs.
I already put forth my requirements for the routing a while ago - there are only two.
a) If two nodes have a direct point to point link, then that link should be the
preferred route.
b) I really want symmetrical routing.
I already know that there are those who believe that symmetrical routing is impossible to
achieve, so maybe I'll give up on (b). I still want (a) though.
7:20 p.m.
On Jan 25, 2016, at 12:33 PM, Johnny Billquist <bqt
at softjar.se> wrote:
On 2016-01-25 18:17, Robert Armstrong wrote:
Symmetrical routing will be impossible. DECnet is not designed in a way that makes it
possible.
Correct.
More precisely, symmetric routing is not a design goal of DECnet. (Nor is it in other
networks I can think of.)
If you set link costs the same at both endpoints, then routes will be symmetric if there
are no equal cost paths between the given source and destination. If equal cost paths
exist, the outcome will depend on the node addresses, and in general you won't get
symmetric routes.
Nor does it matter. There is no end user benefit to symmetric routes, and the debugging
benefits, if any, are small at best.
paul
I will
participate.
Oh, and sure - I'll go along provided I can get the routing results that I want.
Sorry to be selfish, but I'm not going to put up with some nonsensical routing path
just so that we can have standardized costs.
I already put forth my requirements for the routing a while ago - there are only two.
a) If two nodes have a direct point to point link, then that link should be the
preferred route.
b) I really want symmetrical routing.
I already know that there are those who believe that symmetrical routing is impossible to
achieve, so maybe I'll give up on (b). I still want (a) though. 
7:18 p.m.
-----Original Message-----
From: owner-hecnet at Update.UU.SE [mailto:owner-
hecnet at Update.UU.SE] On Behalf Of Johnny Billquist
Sent: 25 January 2016 15:26
To: hecnet <hecnet at Update.UU.SE>
Subject: [HECnet] Circuit costs (again!)
After having my links up and running a couple of days now (and they work
fine), I've come to realize that if we want to have a somewhat efficient
topology, we need to agree on circuit costs.
Now, that said, in todays internet world, network speeds are fast enough
that this isn't really a big problem. So if people don't want to, I'm okay
with
that.
But otherwise we should try and use some agreed upon scheme for setting
costs on circuits. The best suggestion I've seen so far is to base it on ping
times. Would people be willing to adopt this and implement it all over the
place for level 2 routers? Because, if not everyone agrees, then I'm wasting
my time here. :-)
The exact cost scale we can discuss, if we agree on doing it. But it needs to be
pretty much used by everyone, or else it's just a waste of time and energy.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
I am happy to take part, assuming it is still going ahead (I see later discussions in this
thread calling the idea into question).
Regards
Rob
3023
days inactive
3023
days old
19 comments
5 participants
participants (5)
-
bob＠jfcl.com -
bqt＠softjar.se -
mcguire＠neurotica.com -
Paul_Koning＠Dell.com -
robert.jarratt＠ntlworld.com