[HECnet] Fwd: Virtual Cisco North American hub?
Johnny Billquist
bqt at softjar.se
Sun Feb 24 03:22:45 PST 2013
On 2013-02-24 07:44, Ian McLaughlin wrote:
BGP is a distance-vector protocol, DECnet/ISIS is link state,
completely different set of tools.
Agreed. However, DECNET does have a metric - cost - which we can
manipulate in any way we want. We can use it as a proxy for distance.
It doesn't really work. The reason is that you have traffic both ways, and cost is not universal. Packets going in either direction have a different set of costs associated with it, so if you try to use cost to direct traffic to an area, the responses will not adhere to you scheme anyway. But besides, for packets outside your area, you just send to the closest router you know, and that is how far cost goes. (For an endnode it's the designated router, for a level 1 router, it is the closest area router in your area.)
Yes, it is true that we have cost, and also that cost can be used as a proxy for distance. However, when it comes to nodes outside of your area, traffic do not go exactly the way you might naively think.
Your node inside an area will only send the traffic to the "closest" area router, no matter how terrible that node is in respect to getting to the destination area. Also, the area router will just send to the closes area router of the destination area, no matter how lousy that is with regards to the actual location of the destination within that area in respect to the area router you are sending to.
In essence, inter-area routing in DECnet is not optimal, and never will be. It is designed to provide a stable and well functioning layer, but at all any optimal paths.
I'm not sure if a "backbone" of area routers really helps in any way. I guess the most optimial we can do within the constraints of DECnet is to actually perhaps reduce the number of area routers, and then have a full mesh between all area routers. That will probably get us as "optimal" as we can.
Any machine within an area which is not an area router will not be able to select the optimal area router. Any area router, when passing a packet to another area, will not be able to select the optimal area router in the destination area. Those are the simple facts.
If my AS is peering with your AS, I can send you "deaggregated routes
with comunity "do not export" and attach my IGP metrics (and policy)
and send that to you with MED.
In BGP between ASes for econimical reasons you do closest exit to the
peer, so traffic is assumetric, you can listen to MED but if both
sides does it we might be back to assymetric routes.
So in "real world" buisiness polcy is implemented with BGP.
In the real world, I want to get traffic to you off my network as soon
as possible.
If there are several area routers in the destination area, you could in theory want to send to a different one depending on where the end destination lies, but with DECnet that is unknown, so all your traffic will only go towards one destination area router.
If you, by "the real world" imply that you want your packets to get to the destination the most optimal way, it might imply that it should travel longer on your network instead of jumping to another one as soon as possible. It depends on many factors.
With DECnet routing or ISIS L1/L2 you have much less of a tols, and
you can not filter prefixes.
The only thing I was trying to do was to suggest, if we want to make a
backboone of hubs, they better be in the same area.
I don't think anyone is suggesting a backbone (although that's
interesting). All I'm suggesting is a peering point.
Not sure about the backbone thing. But maybe Peter can expand what he is thinking. I'm easy to convince by good ideas.
Not sure what you mean by "peering point".
I saw a bunch of messages in this theread that I think are related to
MTU, the router-tunnel interface has to match the path MTU or things
like routing vectors for high node-numbers will be truncated..
I though that DECnet's maximum packet size was 576? As long as the path
MTU is bigger than this, it shouldn't mangle the DECnet packets, should it?
DECnet's maximum packet size is configurable. It is also negotiated, if I remember right. The both sides can have different setups.
Interfaces have an MTU. That can differ on hops. IP deals with that by allowing packets to be fragmented. It's not something you are aware of above the IP layer (except when you set the Don't Fragment bit, or if you have stupid routers/firewalls in between which can't deal with IP fragments).
So no "mangling" of DECnet packets will happen, even if they exceed the MTU of some hop. The minimum MTU required by IP is something like 200 bytes. It's in the spec somewhere, if someone wants to look it up. (It might be that the minimum MTU is actually something like 64 bytes, but I seem to remember it was something like 200, but I just can't remember for sure right now.)
As for path MTU, I think that concept is only even mentioned in connection with TCP (but I might just be misremembering again). It's used for trying to find the optimal MSS. But TCP deals with streams, and chops it into packets on its own, so in that world, whatever size packets you send is totally irrelevant.
But the concept of pMTU as the lowest MTU of any hops along your path is valid, except that it's hard to actually find out non-intrusively, and exceeding it should not really cost you much anyway. (In fact, if packet loss is insignificant, large fragmented IP packets is better than small non-fragmented ones.)
Johnny
Ian
-p
Ian
On 2013-02-23, at 9:05 PM, Peter Lothberg <roll at Stupi.SE
<mailto:roll at Stupi.SE>> wrote:
Would you care to elaborate? My knowledge of DECnet routing is very =
limited.=3D
=20
It;s no different than I-ISIS.
=20
Within a area you treat all the boarder gateways (are routers) the
same, so for itgoing packets a node inside the area will pick it's
closest area router with external connectivity to senda outgoing
packet through.=20
=20
Another area will pick ANY router it think is usefulle for the
destination area, regardless of the internal topology of that area..=20=
=20
Bottom line:
=20
I have area 59 in Stockholm, Uppsala, Reston VA, and Sunnnyvave CA, if
we want to have a usefull "global core" it needs to be in the same
area....=20
=20
-P
=20
=20
=20
=20
-brian
=20
On Feb 23, 2013, at 19:30, Peter Lothberg <roll at Stupi.SE
<mailto:roll at Stupi.SE>> wrote:
=20
On 02/23/2013 07:21 PM, Cory Smelosky wrote:
I've managed to coax my provider in to giving me another IP
(apparently 5 IPs come with my plan=3D05), so I can get a virtual =
cisco
up on my VPS (in Canada).
=3D20
Brian, would you like to get this all joined in to the mesh? Is a =
NA
hub still needed or has someone else offered one? (This'd be on an
unmetered gigE pipe)
=3D20
I think running a hub on an emulated router maybe isn't the best =
idea.
(sorry to say so)
=3D20
We also have to take in to account how to get some kind of =
resonalble
routing, as the inside of an area is not exchanged with other areas.
=3D20
-P
=20
=20
=20
=20
---
Filter service subscribers can train this email as spam or not-spam =
here: =
http://my.email-as.net/spamham/cgi-bin/learn.pl?messageid=3DE741A17A7E3F11=
E29554144693ED0201
--Apple-Mail=_134C91BE-5BC2-4ABC-B266-7FBB64390094
Content-Transfer-Encoding: quoted-printable
Content-Type: text/html;
charset=us-ascii
<html><head><meta http-equiv=3D"Content-Type" content=3D"text/html =
charset=3Dus-ascii"></head><body style=3D"word-wrap: break-word; =
-webkit-nbsp-mode: space; -webkit-line-break: after-white-space; =
">Peter,<div><br></div><div>If I peered with your three locations, and =
set the cost of each link to be proportional to the distance (or at =
least speed/latency), then my network would prefer the link with the =
lowest cost, but still have the other links as redundant in the case of =
a link failure.</div><div><br></div><div>The routing problems are =
similar to those with BGP. If I'm routing to your AS number, I'm =
going to want to hand it over to the geographically closest point, and =
assume your network is better at connecting all of it's internal parts. =
BGP AS's don't expose the internal structure of the network =
either.</div><div><br></div><div>Ian</div><div><br><div><div>On =
2013-02-23, at 9:05 PM, Peter Lothberg <<a =
href=3D"mailto:roll at Stupi.SE">roll at Stupi.SE</a>> wrote:</div><br =
class=3D"Apple-interchange-newline"><blockquote
type=3D"cite"><blockquote =
type=3D"cite">Would you care to elaborate? My knowledge of DECnet =
routing is very limited.=3D<br></blockquote><br>It;s no different than =
I-ISIS.<br><br>Within a area you treat all the boarder gateways (are =
routers) the<br>same, so for itgoing packets a node inside the area
will =
pick it's<br>closest area router with external connectivity to senda =
outgoing<br>packet through. <br><br>Another area will pick ANY router
it =
think is usefulle for the<br>destination area, regardless of the =
internal topology of that area.. <br><br>Bottom line:<br><br>I have
area =
59 in Stockholm, Uppsala, Reston VA, and Sunnnyvave CA, if<br>we want
to =
have a usefull "global core" it needs to be in the same<br>area.... =
<br><br>-P<br><br><br><blockquote
type=3D"cite"><br><br>-brian<br><br>On =
Feb 23, 2013, at 19:30, Peter Lothberg <<a =
href=3D"mailto:roll at Stupi.SE">roll at Stupi.SE</a>> =
wrote:<br><br><blockquote type=3D"cite"><blockquote type=3D"cite">On =
02/23/2013 07:21 PM, Cory Smelosky wrote:<br><blockquote =
type=3D"cite">I've managed to coax my provider in to giving me another =
IP<br>(apparently 5 IPs come with my plan=3D05), so I can get a virtual =
cisco<br>up on my VPS (in Canada).<br>=3D20<br>Brian, would you like to =
get this all joined in to the mesh? Is a NA<br>hub still needed
or =
has someone else offered one? (This'd be on an<br>unmetered gigE =
pipe)<br></blockquote>=3D20<br> I think running a hub on an emulated =
router maybe isn't the best idea.<br>(sorry to say =
so)<br></blockquote>=3D20<br>We also have to take in to account how to =
get some kind of resonalble<br>routing, as the inside of an area is not =
exchanged with other =
areas.<br>=3D20<br>-P<br></blockquote><br></blockquote><br><br><br>---<br>=
Filter service subscribers can train this email as spam or not-spam =
here: <a =
href=3D"http://my.email-as.net/spamham/cgi-bin/learn.pl?messageid=3DE741A1=
7A7E3F11E29554144693ED0201">http://my.email-as.net/spamham/cgi-bin/learn.p=
l?messageid=3DE741A17A7E3F11E29554144693ED0201</a><br></blockquote></div><=
br></div></body></html>=
--Apple-Mail=_134C91BE-5BC2-4ABC-B266-7FBB64390094--
---
Filter service subscribers can train this email as spam or not-spam
here:
http://my.email-as.net/spamham/cgi-bin/learn.pl?messageid=13F007A27E4A11E2BB17735993ED0201
--
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
More information about the Hecnet-list
mailing list