Talk:Nonblocking minimal spanning switch

Early comments

See Clos network
(preceding unsigned comment by 207.46.71.12 (talk · contribs) 06:09, 19 May 2004)

If any of this makes sense to anybody, please try to explain what is going on. My first question would be: what is an nxm multiplexer? AxelBoldt 21:06 Feb 27, 2003 (UTC)

Surely this should be...

Surely this should be titled "Space Division Switching" or something along those lines!

Cleaning up

I've added the reference (paper by Clos) to the article. (Found without too much difficulty using Google Scholar.)

Finding the reference also allowed additional research into just what this article might be about (since the current article content is not very helpful in that respect). Much of the article is really background for the title subject of the article, not the subject itself. Even so, it does not have enough background (or links to same) to lay an adequate basis for the title subject. For instance, it appears that there should be a description of Clos networks, and definitions given for "strictly nonblocking" and "rearrangeably nonblocking" networks. For useful background in this area see Hunter, Switching Systems and Jajszczyk, 50 Years of Clos Networks, which GScholar turned up.

The actual subject of the article is finally reached in the current so-called "Methods" section, but unfortunately that section is so vague and incomplete as to make, I think, improvement all but impossible for anyone but a subject-area expert (who might be able to find the underlying thought behind the present words, or at least on his own find something useful to say). But that Methods section is central to the article even while being confusingly vague and next to useless. Maybe this article should be tagged with "expert help needed".

-R. S. Shaw 04:11, 9 October 2005 (UTC)

As a semi-expert, I'd appreciate some input on what people would like to know about this subject. When I last worked this article over, I mostly changed the individual sentences into English. There are many Clos network concepts without Wikipedia articles. The distinction between strictly nonblocking (where arbitrary new calls are switchable) and rearrangeably nonblocking (where some new calls require reassignment of nodes currently in use) are much more obscure topics than this algorithm. I'd appreciate it if you changed the section labels to what you think the sections should be. --RSaunders 20:45, 18 October 2005 (UTC)

Okay, but I just stumbled on this article because it was in the cleanup category, and looked it over since it seemed the subject could be interesting. Who knows what the audience in general would want. Anyway, some comments:

1. Thanks for doing some translation into English. I did notice the current article was much improved from the original. I do realize we can't write all the Clos-net-related articles one might want before we fix up this article. We don't need to go into "rearrangeably nonblocking" -- it just occurred to me we were rearranging connections, which meant something to me after I had learned a bit about Clos nets. I changed the "methods" section name to "algorithm", which seemed better to me.
2. Practical implementations of switches

   I find this section interesting, and should be somewhere in Wikipedia, but maybe not in this article. (It maybe should be a core for a Telephone switch article; that article name now is just a redirect to Telephone exchange. The latter article seems like it should be left less technical, since its subject is broader, and so not get this stuff.)

   For this NonblockingMSS article, the problem with this section is that it is too detailed and confusing as background to the article subject. In particular, how the time-division multiplexing works and is equivalent to a crossbar is hard to comprehend, and so distracts in a major way from simply setting necessary background.

3. Problem Space

   Although there is some switch background, a main difficulty when we get to the "Algorithm" section is that the article assumes the reader knows the nature of the realm in which the algorithm is going to be working, but that really hasn't been described. Now that I've personally studied Clos networks in external sources, I think I have the general idea, but on my first reading I sure didn't. From my new perspective I think the single earlier sentence "a switch can be created using a series of layers called subswitches" is pretty much the entire "description" of the problem/solution space, and it surely seems inadequate. There seems to be a lack of words, whether or not there could be a diagram to help.

   I guess the "layers of switches" is the problem space and needs expansion. Isn't it something like a bunch of subswitches which can be thought of as crossbars, organized somehow? Maybe one "layer" of subswitches is connected to inputs and to another layer, then layer to layer, then to a layer that connects to outputs? Are there 3 layers like I guess a Clos network has, or arbitrary number, or what? Is the connection topology between the layers random or structured somehow, or ?

4. The first step in the Nonblocking Minimal Spanning Switch algorithm is to search for a subswitch that contains the needed in and out connections.

   Confusion point: if there are layers like I just wrote of, then a particular subswitch could not be directly ("directly" is implied and meant, isn't it?) connected to both input and output since connections to those would exist only in completely distinct layers of subswitches.

   More confusion: why is this mentioned? It seems like saying "the problem is solved if the problem is solved." Isn't this "search" trivial? Wouldn't one start with the originating (in) line, and immediately have the subswitch it connects to, and then see if that subswitch has a direct (I assume it's direct, but the article doesn't say) connection to the destination (out) line? It's not really a "search" to simply determine if the subswitch in hand has the desired output - it's not looking at any other subswitch or part of the network.

5. If a subswitch with the needed pair of connections can't be found, a pair of subswitches and a connection between them will be needed. One subswitch has a needed input connection. The other has a needed output connection.

   This seems like a somewhat confused statement of the real starting conditions for the algorthm. My guess is that one is given the in and out lines, and trivially from that one has the in and out subswitches (so these are essentially givens for the problem), and the point of the entire algorithm is to find the needed connection between them. Is that right? Seems like the article should actually state what is given and what is being sought through the action of the algorithm.

6. The pair of subswitches' connections can be reorganized with a topological sort.

   Uh, OK, it's introducing a new concept (although if I check the referenced topological sorting article, I don't immediately get help, since it says things like a graph can have "many valid topological sorts" which sounds distressingly vague for current purposes.) Well, I'm ready for further expansion and explanation...

7. The connections of both subswitches are placed in a list that also includes the desired new connection.

   Whew. I wonder exactly what this means. I have a vague idea what placing [some kind (what kind?) of a representation of] a connection into a list might mean, but I'd like some help from the article on this point. Are these list entries including all the intermediate subswitches for each established connection? In one entry big entry per overall connection or a bunch of entries? But then the text gets really murky with "includes the desired new connection" - if you have the connection already, wouldn't you be done? The point the algoritm is to find the/a connection, and we haven't yet, so you can't yet list it!

8. topological sort??

   And now we've reached the end of the paragraph on the topological sort, and still have close to zero idea what it is or how it forms a part of the NMSS algorithm. The next and final paragraph has very, very unclear relationship to this one - is that paragraph's actions somehow part of the topological sort, or is still preparing more data for the topological sort, or is it done after the topological sort? Are these two paragraphs even related to one another or are they talking about completely different subjects?

9. Connections are traced through the list by the Nonblocking Minimal Spanning Switch algorithm.

   Since we don't know the nature of the contents of this list, figuring out how tracing could be done using it is mysterious.

10. Starting from some input or output, ...

    Why is this so vague? Is it trying to say we're avoiding starting with some other kind of entry that might be in the list? (Again, what kind of entries might there be?) Why can't it just start with, say, the input we were given to establish a new connection for? Just more puzzlement.

11. ...the software traces a connection to an output, ...

    Uh, even if we started with an output, as was a stated possibility??? puzzle, puzzle

12. ...then traces the other connection at that output to an input, and so forth, until it comes to an end.

    Huh? What does "comes to an end" mean? Apparently an output or an input isn't an end (we kept tracing), so what is an end?

13. Each time it traces from input to output, the connection is placed in one subswitch, and removed from the list.

    Uh, what is this operation "connection is placed in one subswitch"? How is it "placed"? Is this placing into another list of some kind? And in what subswitch?? An arbitrary one? Doesn't the subswitch have to be carefully chosen in some manner? I suppose the phrase "removed from the list" would make sense at this point, if we only had previously gotten a good idea of what this list looks like.

14. When it traces from output to input, the connection is placed in the other subswitch"

    Umm, what "other" subswitch? Aren't there many subswitches under consideration?

15. To complete correctly, tracing must begin with single connection inputs and outputs, and only then trace double-ended inputs and outputs, which might form loops.

    This has a plausible ring to it, but I'm so lost by this time I can only ignore it. Also, I don't remember the article even hinting at what "double-ended inputs and outputs" might be.

16. Telephone exchange#Topological sort

    This other article has a section which seems to almost verbatim copy the original version of the Algorithm section of this NMSS article. If this article gets improved, maybe something (like replacement with a link) should be done with that section.

Yours for article improvement -R. S. Shaw 23:25, 19 October 2005 (UTC)

As a semi-expert :-), I could clean this up, but I think most of the article needs to be cut, because it does not directly relate to the type of switch in the title. The article reads more like an overview of switching technology. I agree that much of what is here could be useful in other articles, but I'm not in a position to create several articles out of one. Elasticsoul 22:24, 26 October 2005 (UTC)

I'm a domain expert. I actually have written the software to control a small three-layer switch using the Clos algorithm, and read some of the research on it, and its application. Further, the text resembles some text that I originally wrote for telephone switch several years ago. I noticed when it migrated to telephone exchange, but haven't followed it much since. I'm terribly sorry that it seemed unclear. I just now tried to add lots and lots of context to the algorithm and explain it better. And maybe it does belong in telephone switch, but if you put it there, the British are quite likely to move it to telephone exchange, which some of them believe is an exact cognate (they do not seem to distinguish the machine from the area code). Best wishes User:Ray Van De Walker

Diagram of rerouting

I added a diagram of rerouting a NMSS so additional signals can be routed. (Actually, 2 diagrams). Let me know if there are any comments on how they came out. RJFJR 21:48, 13 May 2007 (UTC)

Geometric series?

However, the crossbar switch does so at the expense of using N² (N squared) simple SPST switches. For large N (and the practical requirements of a phone switch are considered large) this geometric growth was prohibitively expensive.

What has the geometric series to do with this? Thanks, --Abdull (talk) 09:03, 4 August 2008 (UTC)

Nothing. It should be geometric growth, I think. Ray Van De Walker 23:14, 13 December 2009 (UTC)