Cost don't impact performance. .. ....
Of course not, but two hops via the bridge (as compared to one hop via
Multinet) does...
Bob
But 10 or 20 on SG1 for the multinet link to you would make you in control
over all of it. You could chise your bridge, my Multinet or the SG1 multinet.
I can give you login on STUPI...
-P
Would it be possible to change the metric on SG1 on TCP-0-2 to for example
10?
Actually, what hapen if we set all multinet UDP links to 20 and TCP links
to 10
10 and 20 are way too high. The problem is that the default cost for QNA
and UNA circuits is 3 (or 4 - I don't remember). Everybody connected to
Johnny's bridge therefore sees a cost of 3 to everybody else on the bridge,
even though that connection actually requires a hop over to Sweden (psilo)
and back again. Any two machines with Johnny's bridge would never use the
Multinet connection, even if it were a direct link between the two.
Cost don't impact performance. .. ....
I guess that the "problem" here is that most people have local
machines on the same segent that is bridged.
Having a "to low" metric makes it impoissible to prefer a better path
that have serveral hops, over the bridge.....
This sounds strange but.....
Reliable - PTP link 10
Unreliable PTP link 20
Ethernet bridged 30
Would most likely render the optimal performance.
-P
Would it be possible to change the metric on SG1 on TCP-0-2 to for example
10?
Actually, what hapen if we set all multinet UDP links to 20 and TCP links
to 10
10 and 20 are way too high. The problem is that the default cost for QNA
and UNA circuits is 3 (or 4 - I don't remember). Everybody connected to
Johnny's bridge therefore sees a cost of 3 to everybody else on the bridge,
even though that connection actually requires a hop over to Sweden (psilo)
and back again. Any two machines with Johnny's bridge would never use the
Multinet connection, even if it were a direct link between the two.
Bob
Bob and I are experimenting with Multinet in TCP mode, seems fine, but
I would like to se some traffic on the link, and it's assymetric.
NCP>tell legato sh node mim
Node Volatile Summary as of 30-MAY-2014 02:13:14
Node State Active Delay Circuit Next node
Links
1.13 (MIM) TCP-5-9 59.58 (STUPI)
NCP>tell mim sh node legato
Node Volatile Summary as of 30-MAY-2014 02:13:36
Node State Active Delay Circuit Next node
Links
2.1 (LEGATO) 0 6 19.41 (SG1)
NCP>tell sg1 sh cir TCP-0-2 char
Circuit Volatile Characteristics as of 30-MAY-2014 02:18:56
Circuit = TCP-0-2
State = on
Service = enabled
Cost = 1
Hello timer = 300
Verification = disabled
Adjacent node = 2.1 (LEGATO)
Listen timer = 600
Would it be possible to change the metric on SG1 on TCP-0-2 to for
example 10?
Actually, what hapen if we set all multinet UDP links to 20
and TCP links to 10 ? Things would then be symetric and take the most
reliable path.... or?
--P
On 2014-05-30 01:52, Cory Smelosky wrote:
On Fri, 30 May 2014, Johnny Billquist wrote:
It should already be INStalled. That don't really take any memory.
It's normally the running of tasks that takes memory.
The nodename server is managed by a program called SCP. To start it,
you do "SCP START".
NETINS asks me to start NNS. If I say YES, NCP START SYS says GEN is
too fragmented. If I say NO and fiddle with the pool it comes up fine.
Ah. The memory is too fragmented because NT.NNS is started before the main network stuff. Yes, changing the order is a good idea here. But it's not the installing that is the problem, but the fact that the task is actually started before the network.
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 Fri, 30 May 2014, Johnny Billquist wrote:
It should already be INStalled. That don't really take any memory. It's normally the running of tasks that takes memory.
The nodename server is managed by a program called SCP. To start it, you do "SCP START".
NETINS asks me to start NNS. If I say YES, NCP START SYS says GEN is too fragmented. If I say NO and fiddle with the pool it comes up fine.
Johnny
--
Cory Smelosky
http://gewt.net Personal stuff
http://gimme-sympathy.org Projects
On 2014-05-30 01:47, Cory Smelosky wrote:
On Fri, 30 May 2014, Johnny Billquist wrote:
On 2014-05-30 01:42, Cory Smelosky wrote:
On Fri, 30 May 2014, Johnny Billquist wrote:
I can't define remote nodes under RSX though. Which means copying
doesn't work.
Huh? What do you mean by can't define?
I can't SET NODE <number> NAME <name>. It gives some sort of
syntax/invalid definition error. I tried various variations and followed
the HELP exactly.
It should work. Do you know if the name service is running?
There should be a task called NT.NNS running on the system. If that is
not running, then I suspect you might have problems with nodenames.
Yeah. That's not running. Not enough memory. Now that I know what
it's called I might be able to manually INSert it, though.
It should already be INStalled. That don't really take any memory. It's normally the running of tasks that takes memory.
The nodename server is managed by a program called SCP. To start it, you do "SCP START".
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 2014-05-29 20:03, Jordi Guillaumes i Pons wrote:
El 29/05/2014, a les 19.59, Cory Smelosky <b4 at gewt.net> va escriure:
El 29/05/2014, a les 19.56, Cory Smelosky <b4 at gewt.net> va escriure:
Can you give an example? I assume you are using CFE to do the definitions.
I was using NCP. I had no idea how to define a node any other way due to the fact I have very minimal RAM.
With NCP you just can add nodes to the volatile database (SET NODE xxx). You need to use a program named CFERES which lives in [5,54] to do permanent definitions. IIRC you must SET /UIC=[5,54], then RUN CFERES and answer CETAB.MAC to the filename prompt. Then you will get a CFE> prompt and from then you can add your node definitions. (DEFINE NODE xxx). To activate the changes you must reboot RSX (Or perhaps stopping and starting the network is enough, but I'm not sure).
Oh, IIRC you can use SET NODE xxxx in CFE, but it acts as a DEFINE NODE anyway.
Right, NCP is only for the volatile database. CFE is for the permanent one.
CFE or CFERES or CFEFSL. Either does the job, they have just been task built in different ways...
You do not have to set your UIC in order to run it, but you probably needs to be under a system UIC, as well as having privileges.
When adding node names you do not need to reboot, or even restart the network. Nodename changes take effect immediately.
Other type of changes might require either a network restart or a reboot, depending on what you are changing.
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 Fri, 30 May 2014, Johnny Billquist wrote:
On 2014-05-30 01:42, Cory Smelosky wrote:
On Fri, 30 May 2014, Johnny Billquist wrote:
I can't define remote nodes under RSX though. Which means copying
doesn't work.
Huh? What do you mean by can't define?
I can't SET NODE <number> NAME <name>. It gives some sort of
syntax/invalid definition error. I tried various variations and followed
the HELP exactly.
It should work. Do you know if the name service is running?
There should be a task called NT.NNS running on the system. If that is not running, then I suspect you might have problems with nodenames.
Yeah. That's not running. Not enough memory. Now that I know what it's called I might be able to manually INSert it, though.
Johnny
--
Cory Smelosky
http://gewt.net Personal stuff
http://gimme-sympathy.org Projects
