On 1/12/2013 8:17 AM, Cory Smelosky wrote:
On 12 Jan 2013, at 08:16, Brian Hechinger<wonko at 4amlunch.net> wrote:
>You need the new walking code. I'll send an email when I'm at my computer next.
Can I have a copy as well? I won't do anything with it, but i'm just curious.
Here's the current version. This won't be the final version as it will need things like Cisco support added.
It dumps the data to pickle files. To read them just do something like this:
def load_data():
# Open pickle files
node_char_pkl = open('data/node_char.pkl', 'rb')
known_circ_pkl = open('data/known_circ.pkl', 'rb')
adj_nodes_pkl = open('data/adj_nodes.pkl', 'rb')
broken_nodes_pkl = open('data/broken_nodes.pkl', 'rb')
# Load pickled data
node_char = pickle.load(node_char_pkl)
known_circ = pickle.load(known_circ_pkl)
adj_nodes = pickle.load(adj_nodes_pkl)
broken_nodes = pickle.load(broken_nodes_pkl)
# Close pickle files
node_char_pkl.close()
known_circ_pkl.close()
adj_nodes_pkl.close()
broken_nodes_pkl.close()
#def load_data()
Data layout is as follows:
broken_nodes = set of nodes that doesn't answer to NCP
node_char = dictionary of dictionaries of the "show exec char" output. Primary index is node address. Secondary index is whatever is on the left hand side of the = in the output.
ex: node_char['1.13']['Identification'] = "RSX system at Update, Sweden"
one exception to this is node_char[address]['node_name'] which is the name as parsed out of the "Executor node" line.
known_circ = dictionary of dictionaries. Primary index is node queried, secondary index is circuit address.
adj_nodes = dictionary of dictionaries. Primary index is node queried, secondary index is adj node.
-brian
On 2013-01-12 17:32, Bob Armstrong wrote:
Johnny Billquist wrote:
Take WXP for example...
What is the story with WXP ??? It responds to NCP SHOW EXEC CHAR with a
privilege violation error. That's a bit odd...
Not really.
Any DECnet node that don't have a default account set up, don't allow random access through proxy, and which don't allow connections to objects without authorization will do the same. I can easily cause RSX to give the same error to you.
Anyway, WXP is my Windows machine, and since giving access to NICE also implies giving access to FAL, and since Windows is so utterly stupid that this means access to all of the disk, I'm not going to let random people get access to WXP. But I play with it myself...
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 2013-01-12 16:38, sampsa at mac.com wrote:
OK what do you suggest?
Ah. Now, that is the trickier question. The only working solution I can think of is to have the information manually provided.
In order to not put too much burden on one individual, it would be easier if we had an agreed on format, and then a bunch of known places where to fetch those files, and then aggregate it. The aggregation can be done in multiple places, if wanted. Once we have the basic structure in place, we can deal with this in all kind of ways.
I'm not too fond of the solution, since it is bound to be out of date sooner or later, but I don't see any other solution that is any better. But I'm also open to suggestions...
Johnny
sampsa
On 12 Jan 2013, at 17:33, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 16:30, sampsa at mac.com wrote:
On 12 Jan 2013, at 17:28, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 16:23, sampsa at mac.com wrote:
I already do SHOW ADJ NODES on the router nodes, so that should have most non-CISCO stuff covered no?
But they show up as gazillion ptp links unless you somehow start to figure out that this is actually a bus. And then (not that I'm aware of any such links on HECnet right now) the multidrop links, which connects several machines, but which actually are lots of ptp connections...
Again NCP can't tell me that - a simple bit of metainfo in FAL could :)
Yes. Which is my point. NCP falls severely short if you want to use it for mapping. Something else is needed, so let's do that something else instead, and do that properly.
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
--
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
Johnny Billquist wrote:
Take WXP for example...
What is the story with WXP ??? It responds to NCP SHOW EXEC CHAR with a
privilege violation error. That's a bit odd...
Bob
On 2013-01-12 16:30, sampsa at mac.com wrote:
On 12 Jan 2013, at 17:28, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 16:23, sampsa at mac.com wrote:
I already do SHOW ADJ NODES on the router nodes, so that should have most non-CISCO stuff covered no?
But they show up as gazillion ptp links unless you somehow start to figure out that this is actually a bus. And then (not that I'm aware of any such links on HECnet right now) the multidrop links, which connects several machines, but which actually are lots of ptp connections...
Again NCP can't tell me that - a simple bit of metainfo in FAL could :)
Yes. Which is my point. NCP falls severely short if you want to use it for mapping. Something else is needed, so let's do that something else instead, and do that properly.
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 2013-01-12 16:26, Peter Lothberg wrote:
On 2013-01-12 16:10, Peter Lothberg wrote:
If you ask me, I think this idea is dead. You cannot get a good
representation of how the topology really is from trying to walk nodes
using NCP.
(remove bridged ethernet that are not point-to-point with only 2
nodes).
Walk through the "known cir" and then "cir char" and build your own
connectivity tree with the metrics.
Plot all links
Apply DECnet routing rules and Colour the links that carry traffic
with the current metrics and/up/down condition, use two colours, one
for transmitt one for receive (as links can end up being simplex).
Well, it all depends on what you want to know. If you want to properly
understand the topology you also want to represent the topology of
bridged ethernet segments, as well as ethernets in general. NCP don't
have any clue about this.
Sure, if you want to only have Multinet ptp links, then it's easy. But
that would a rather severe restriction on the network technology.
We have ethernets (both bridged and local), as well as Cisco tunnels.
Both of which walking with NCP fails on.
NCP don't fail on DECboxes on ethernets.
Right. Sloppy working on my part.
The plotting thing would do the right thing even on the ethernet, even
with the ethernet it ended up being a simplex path for some
multi-homed sites.
There are two ways of represnting the the Ethernet,
1 You draw it as a bus ignore that the links to it have
different characteristics.
Right, except I didn't follow your comment about characteristics. Are you talking about the fact that different machines have different parameters for the circuits? Or that when you have a bridge, you get very different characteristics if you talk to something on the same segment as compared to something that needs to go over the bridge?
2 You drow it as all the point-to-point links it represents.
Look how IP handles Ethernet, it builds simplex TX only point
to point links using ARP and a dest-ip address to MAC address
mapping.
Drawing all as ptp links is both ugly, and I'd like to say incorrect.
Drawing it as a bus is what I thing should be done.
But there is yet another complication. How do you figure this out? After all, there also exists multidrop links in DECnet, which do connect several machines, but which really are ptp links.
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 2013-01-12 16:23, sampsa at mac.com wrote:
I already do SHOW ADJ NODES on the router nodes, so that should have most non-CISCO stuff covered no?
But they show up as gazillion ptp links unless you somehow start to figure out that this is actually a bus. And then (not that I'm aware of any such links on HECnet right now) the multidrop links, which connects several machines, but which actually are lots of ptp connections...
Johnny
sampsa
On 12 Jan 2013, at 17:21, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 16:11, sampsa at mac.com wrote:
I won't - which is why I thought the bridge users could add some meta-info to their INFO.TXTs to help me map how they are connected, for example.
And all other ethernets?
Johnny
sampsa
On 12 Jan 2013, at 17:02, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 15:53, sampsa at mac.com wrote:
Johnny,
That's a graphic error, not a data collection one..
Ok. Since I don't know details on your implementation I'll have to take your word on that.
However, I still totally fail to see how you will ever be able to get the required information out from NCP, since NCP don't know (apart from not all nodes even talking NICE).
Johnny
sampsa
On 12 Jan 2013, at 16:35, Johnny Billquist <bqt at softjar.se> wrote:
Yes. That is one additional complication. However, looking at the current output from sampsa, it looks as if ethernet and multinet links are mixed on the same segment as well.
--
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
--
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
--
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 2013-01-12 16:10, Peter Lothberg wrote:
If you ask me, I think this idea is dead. You cannot get a good
representation of how the topology really is from trying to walk nodes
using NCP.
(remove bridged ethernet that are not point-to-point with only 2
nodes).
Walk through the "known cir" and then "cir char" and build your own
connectivity tree with the metrics.
Plot all links
Apply DECnet routing rules and Colour the links that carry traffic
with the current metrics and/up/down condition, use two colours, one
for transmitt one for receive (as links can end up being simplex).
Well, it all depends on what you want to know. If you want to properly
understand the topology you also want to represent the topology of
bridged ethernet segments, as well as ethernets in general. NCP don't
have any clue about this.
Sure, if you want to only have Multinet ptp links, then it's easy. But
that would a rather severe restriction on the network technology.
We have ethernets (both bridged and local), as well as Cisco tunnels.
Both of which walking with NCP fails on.
NCP don't fail on DECboxes on ethernets.
The plotting thing would do the right thing even on the ethernet, even
with the ethernet it ended up being a simplex path for some
multi-homed sites.
There are two ways of represnting the the Ethernet,
1 You draw it as a bus ignore that the links to it have
different characteristics.
2 You drow it as all the point-to-point links it represents.
Look how IP handles Ethernet, it builds simplex TX only point
to point links using ARP and a dest-ip address to MAC address
mapping.
-P
On 2013-01-12 16:21, Johnny Billquist wrote:
On 2013-01-12 16:10, Peter Lothberg wrote:
If you ask me, I think this idea is dead. You cannot get a good
representation of how the topology really is from trying to walk nodes
using NCP.
(remove bridged ethernet that are not point-to-point with only 2
nodes).
Walk through the "known cir" and then "cir char" and build your own
connectivity tree with the metrics.
Plot all links
Apply DECnet routing rules and Colour the links that carry traffic
with the current metrics and/up/down condition, use two colours, one
for transmitt one for receive (as links can end up being simplex).
Well, it all depends on what you want to know. If you want to properly
understand the topology you also want to represent the topology of
bridged ethernet segments, as well as ethernets in general. NCP don't
have any clue about this.
Sure, if you want to only have Multinet ptp links, then it's easy. But
that would a rather severe restriction on the network technology.
We have ethernets (both bridged and local), as well as Cisco tunnels.
Both of which walking with NCP fails on.
By the way. One thing I have not even touched on yet is that another "restriction" in all this is that not all nodes who even talk NICE will allow random other machines to query them.
Take WXP for example...
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
Brian H is working on the CISCO stuff and maybe the bridge users could provide some meta-info about their connections say in a BRIDGE.INFO file or something?
sampsa
On 12 Jan 2013, at 17:21, Johnny Billquist <bqt at softjar.se> wrote:
On 2013-01-12 16:10, Peter Lothberg wrote:
If you ask me, I think this idea is dead. You cannot get a good
representation of how the topology really is from trying to walk nodes
using NCP.
(remove bridged ethernet that are not point-to-point with only 2
nodes).
Walk through the "known cir" and then "cir char" and build your own
connectivity tree with the metrics.
Plot all links
Apply DECnet routing rules and Colour the links that carry traffic
with the current metrics and/up/down condition, use two colours, one
for transmitt one for receive (as links can end up being simplex).
Well, it all depends on what you want to know. If you want to properly understand the topology you also want to represent the topology of bridged ethernet segments, as well as ethernets in general. NCP don't have any clue about this.
Sure, if you want to only have Multinet ptp links, then it's easy. But that would a rather severe restriction on the network technology.
We have ethernets (both bridged and local), as well as Cisco tunnels. Both of which walking with NCP fails on.
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
