McLaughlin <ian at platinum.net> va escriure:
I'd just like to say Thank You to everyone on the network for giving us all this huge plaything that we all dreamed of playing with a quarter of a century ago :)
I strongly second this!! The awesomeness of all of you is beyond my ability to express in English!
On 9 Jan 2013, at 12:04, Ian McLaughlin <ian at platinum.net> wrote:
On 2013-01-09, at 8:44 AM, Johnny Billquist <bqt at softjar.se> wrote:
A big part of the general problem is that all these numbers are local for each end of connections, so that you can get very asymetrical setups.
Also, the metrics are set on the circuit, and not individual destinations. While this worked more or less ok for DEC back in the day, with the bridge, the cost of two different destinations over the same circuit could in reality be very different.
In the end, it boils down to a couple of things for tweaking the costs.
1. Preference of tunnel or bridge. If you prefer using the bridge, then just set the ethernet circuit cost lower than the tunnel. The tunnel can be either a multinet tunnel, or a tunnel on the cisco.
2. If the rest of your network isn't on the same ethernet as the bridge is, then you can use tweaks and control to your hearts content.
3. If the rest of your network don't have some other path out, then it don't matter if the ethernet circuit cost is high, so still feel free to tweak it as needed.
I think that you should be able to tweak the costs freely almost all the time, without any issues. However, you will sometimes not get the results you are looking for if the other end, and all intermediate nodes have also set their costs based on similar views as yours.
Johnny
If this was a production network with large volumes of data travelling across it, careful selection of metrics would be very important for performance. However, what we have is an ad-hoc network with islands controlled by individuals. These individuals have differing skill levels, hardware collections, and requirements from the network. There is no 'backbone' provider or administrator, so the metrics end up being all over the place.
I would like to identify what results in me having rather low FAL access speeds, however. I don't know if my network is misconfigured, or if it's just a result of using real hardware on a remote node.
In my case, if I had a link to a US area, who also had a link to a second US area, I would want to take the path from the first to the second, instead of going through Sweden and back (although I'm sure Sweden is beautiful this time of year). Assigning a fixed cost to the path type doesn't take in to account the real latency of that path.
What would be ideal is if the mapping project that is currently underway could show the metrics of the links (in both directions, because they could be assymetrical). If the map could also show relative geographical
locations, that would also be a good proxy for latency estimates on links.
That would be quite useful.
But you know what? This is the fun part of Hecnet. None of us are doing this because it's a mission-critical network (for some of us, that's probably what our day jobs are anyway) - we're doing it because it's fun, and we can play with 'what-if' without breaking anything important.
What are you talking about? HECnet is mission critical. ;)
Sometimes the tone in here gets a little serious. I'd just like to say Thank You to everyone on the network for giving us all this huge plaything that we all dreamed of playing with a quarter of a century ago :)
I'd like to thank everyone here as well, you've helped me learn a lot and given me something fun to play with. :)
Ian
--
Cory Smelosky
http://gewt.net/ Personal stuff!
http://gimme-sympathy.org/ My permanently-a-work-in-progress pet project.
On 2013-01-09, at 8:44 AM, Johnny Billquist <bqt at softjar.se> wrote:
A big part of the general problem is that all these numbers are local for each end of connections, so that you can get very asymetrical setups.
Also, the metrics are set on the circuit, and not individual destinations. While this worked more or less ok for DEC back in the day, with the bridge, the cost of two different destinations over the same circuit could in reality be very different.
In the end, it boils down to a couple of things for tweaking the costs.
1. Preference of tunnel or bridge. If you prefer using the bridge, then just set the ethernet circuit cost lower than the tunnel. The tunnel can be either a multinet tunnel, or a tunnel on the cisco.
2. If the rest of your network isn't on the same ethernet as the bridge is, then you can use tweaks and control to your hearts content.
3. If the rest of your network don't have some other path out, then it don't matter if the ethernet circuit cost is high, so still feel free to tweak it as needed.
I think that you should be able to tweak the costs freely almost all the time, without any issues. However, you will sometimes not get the results you are looking for if the other end, and all intermediate nodes have also set their costs based on similar views as yours.
Johnny
If this was a production network with large volumes of data travelling across it, careful selection of metrics would be very important for performance. However, what we have is an ad-hoc network with islands controlled by individuals. These individuals have differing skill levels, hardware collections, and requirements from the network. There is no 'backbone' provider or administrator, so the metrics end up being all over the place.
In my case, if I had a link to a US area, who also had a link to a second US area, I would want to take the path from the first to the second, instead of going through Sweden and back (although I'm sure Sweden is beautiful this time of year). Assigning a fixed cost to the path type doesn't take in to account the real latency of that path.
What would be ideal is if the mapping project that is currently underway could show the metrics of the links (in both directions, because they could be assymetrical). If the map could also show relative geographical locations, that would also be a good proxy for latency estimates on links.
But you know what? This is the fun part of Hecnet. None of us are doing this because it's a mission-critical network (for some of us, that's probably what our day jobs are anyway) - we're doing it because it's fun, and we can play with 'what-if' without breaking anything important.
Sometimes the tone in here gets a little serious. I'd just like to say Thank You to everyone on the network for giving us all this huge plaything that we all dreamed of playing with a quarter of a century ago :)
Ian
On 1/9/2013 11:44 AM, Johnny Billquist wrote:
On 2013-01-09 17:19, Peter Lothberg wrote:
On 1/9/2013 1:59 PM, Peter Lothberg wrote:
On 1/8/2013 9:01 PM, Peter Lothberg wrote:
Set up a (cisco) tunnel to 130.238.19.60 (and tell me your side IP
address).
I think it's time to come up with something a little better to track all
these.
Thoughts?
Build a tool that takes requests and generates configs (full mesh) and
pushes them to the routers.
(maybe we can have the router configs in a wiki?)
Great mind think alike. :)
I had though pretty much exactly that.
If we are going to do that, however, we should decide on a range of
tunnels to use.
For example on my router, 0, 1 and 2 are DECnet tunnels, 3 is my IPv6
tunnel to HE and 4 is a DECnet tunnel.
Re-pushing the configs in this case would be ugly, so I think it would
be wise to say, tunnels X through Y are dedicated to auto-DECnet tunnels
so I can write cisco configs to deal with that appropriately.
I will write something. :)
-brian
If you knew the capability (what kind of tunnels) a node can support and you have
a fixed priority for in what order to chose, given two nodes you can automatically
determine the required tunnel type and don't need to encode it in to you global
description as it's a pair-wise local matter..... -:)
A big part of the general problem is that all these numbers are local for each end of connections, so that you can get very asymetrical setups.
Also, the metrics are set on the circuit, and not individual destinations. While this worked more or less ok for DEC back in the day, with the bridge, the cost of two different destinations over the same circuit could in reality be very different.
In the end, it boils down to a couple of things for tweaking the costs.
1. Preference of tunnel or bridge. If you prefer using the bridge, then just set the ethernet circuit cost lower than the tunnel. The tunnel can be either a multinet tunnel, or a tunnel on the cisco.
2. If the rest of your network isn't on the same ethernet as the bridge is, then you can use tweaks and control to your hearts content.
3. If the rest of your network don't have some other path out, then it don't matter if the ethernet circuit cost is high, so still feel free to tweak it as needed.
I think that you should be able to tweak the costs freely almost all the time, without any issues. However, you will sometimes not get the results you are looking for if the other end, and all intermediate nodes have also set their costs based on similar views as yours.
I was referring to the Cisco tunnel interface numbering, not path cost.
You did address a question I had about that, however. :)
I can make it configurable, but ultimately it might be easiest to choose a path cost that is most likely going to work for everyone.
-brian
-brian
On 2013-01-09 17:19, Peter Lothberg wrote:
On 1/9/2013 1:59 PM, Peter Lothberg wrote:
On 1/8/2013 9:01 PM, Peter Lothberg wrote:
Set up a (cisco) tunnel to 130.238.19.60 (and tell me your side IP
address).
I think it's time to come up with something a little better to track all
these.
Thoughts?
Build a tool that takes requests and generates configs (full mesh) and
pushes them to the routers.
(maybe we can have the router configs in a wiki?)
Great mind think alike. :)
I had though pretty much exactly that.
If we are going to do that, however, we should decide on a range of
tunnels to use.
For example on my router, 0, 1 and 2 are DECnet tunnels, 3 is my IPv6
tunnel to HE and 4 is a DECnet tunnel.
Re-pushing the configs in this case would be ugly, so I think it would
be wise to say, tunnels X through Y are dedicated to auto-DECnet tunnels
so I can write cisco configs to deal with that appropriately.
I will write something. :)
-brian
If you knew the capability (what kind of tunnels) a node can support and you have
a fixed priority for in what order to chose, given two nodes you can automatically
determine the required tunnel type and don't need to encode it in to you global
description as it's a pair-wise local matter..... -:)
A big part of the general problem is that all these numbers are local for each end of connections, so that you can get very asymetrical setups.
Also, the metrics are set on the circuit, and not individual destinations. While this worked more or less ok for DEC back in the day, with the bridge, the cost of two different destinations over the same circuit could in reality be very different.
In the end, it boils down to a couple of things for tweaking the costs.
1. Preference of tunnel or bridge. If you prefer using the bridge, then just set the ethernet circuit cost lower than the tunnel. The tunnel can be either a multinet tunnel, or a tunnel on the cisco.
2. If the rest of your network isn't on the same ethernet as the bridge is, then you can use tweaks and control to your hearts content.
3. If the rest of your network don't have some other path out, then it don't matter if the ethernet circuit cost is high, so still feel free to tweak it as needed.
I think that you should be able to tweak the costs freely almost all the time, without any issues. However, you will sometimes not get the results you are looking for if the other end, and all intermediate nodes have also set their costs based on similar views as yours.
Johnny
On 1/9/2013 1:59 PM, Peter Lothberg wrote:
On 1/8/2013 9:01 PM, Peter Lothberg wrote:
Set up a (cisco) tunnel to 130.238.19.60 (and tell me your side IP
address).
I think it's time to come up with something a little better to track all
these.
Thoughts?
Build a tool that takes requests and generates configs (full mesh) and
pushes them to the routers.
(maybe we can have the router configs in a wiki?)
Great mind think alike. :)
I had though pretty much exactly that.
If we are going to do that, however, we should decide on a range of
tunnels to use.
For example on my router, 0, 1 and 2 are DECnet tunnels, 3 is my IPv6
tunnel to HE and 4 is a DECnet tunnel.
Re-pushing the configs in this case would be ugly, so I think it would
be wise to say, tunnels X through Y are dedicated to auto-DECnet tunnels
so I can write cisco configs to deal with that appropriately.
I will write something. :)
-brian
If you knew the capability (what kind of tunnels) a node can support and you have
a fixed priority for in what order to chose, given two nodes you can automatically
determine the required tunnel type and don't need to encode it in to you global
description as it's a pair-wise local matter..... -:)
-P
On 2013-01-09 14:01, Steve Davidson wrote:
-----Original Message-----
From: owner-hecnet at Update.UU.SE
[mailto:owner-hecnet at Update.UU.SE] On Behalf Of Peter Lothberg
Sent: Wednesday, January 09, 2013 08:49
To: hecnet at Update.UU.SE
Cc: hecnet at Update.UU.SE
Subject: Re: [HECnet] HECnet performance....
Peter,
Everything is now set up on my end. I now have 3 Cisco
tunnels. The
= metric is 10 on all of them.
Change the metrics to match my side:
Tunnel2 Cost 10
tunnel to 192.108.200.213 Cost 20
tunnel to 199.0.131.2 Cost 20
The majority of areas now route through you according to my router:
Given that we have no way of setting metrics on the "bridged ethernet"
links, this is the best we can do that works in the general case.
I'm now trying to get the Multinet links to
192.108.200.211 59.58 (STUPI) cleaned up and re_established,
right now only the link to Legato is working properly. SG1
had cost 1 on their side instead of 10.
I should add that if SG1:: and LEGATO:: need to pass information, the
circuit costs favor the Multinet Tunnel links between the two major US
hubs. I would rather the Multinet Tunnel be used instead of using the
bridge link to Sweden and then back again.
I should probably just point out that there is no reason to set up the bridges to all be just connected to Update, and have everything travel through that one hub. The bridge works equally well with multiple hops. Although it does not work with a mesh, since it don't do any kind of spanning tree.
Johnny
On 1/9/2013 2:00 PM, Peter Lothberg wrote:
On 01/08/2013 05:38 PM, Brian Hechinger wrote:
Set up a (cisco) tunnel to 130.238.19.60 (and tell me your side IP
address).
I think it's time to come up with something a little better to track all
these.
Thoughts?
A subdomain under some related domain (perhaps
<sub>.hecnet.update.uu.se?) with the tunnel endpoints as A records?
-Dave
We are not limited to route DECnet packets, we could turn on IP on the
tunnels, and CLNP/Phase5..
If it comes down to needing to use SNMP I'd much prefer this than exposing SNMP over the internet.
-brian
On 1/8/2013 5:44 PM, Dave McGuire wrote:
On 01/08/2013 05:38 PM, Brian Hechinger wrote:
Set up a (cisco) tunnel to 130.238.19.60 (and tell me your side IP
address).
I think it's time to come up with something a little better to track all
these.
Thoughts?
A subdomain under some related domain (perhaps
<sub>.hecnet.update.uu.se?) with the tunnel endpoints as A records?
The problem here is that it isn't searchable. You need to know the A record name. It would be more useful for me to have it in a db.
-brian
