List:General Discussion« Previous MessageNext Message »
From:shawn wilson Date:August 23 2012 2:42pm
Subject:Re: Machine Learning
View as plain text  
just a few thoughts (things to look into)

if you want to populate a db with command parameters, i'd mine the man
pages. it's a consistent format and you should be able to find a
parser for whatever language you prefer.

if you want to see what has been entered and statistics about that
process, look at the audit framework. it won't tell you success or
failure, but you might be able to derive that based on the time,
memory, and process usage that you can get from audit. auditd also
maintains a db - you might look into the engine and schema they use.

you might also find some interesting topics looking into language
processing (after all, commands are a type of language). people
generally use other database engines to do this type of thing -
elastic search or some other things people have done with lucene
maybe.

do let us know where you go with this. it sounds quite interesting
(and might help me with something i've been looking into)... and
getting zsh to autocomplete options might be a bonus if you parse man
pages and shove it into a db (i might do that part for you if i get a
lazy afternoon).

On Wed, Aug 22, 2012 at 7:51 PM, Garot Conklin <garotconklin@stripped> wrote:
> I am going to work up a visual diagram for this to better... visualize (LOL) the lay
> of the land. I will forward it when i have it completed then I hope we can continue the
> architectural discussion here!  Thanks
>
> -Garot
>
>
> garotconklin@stripped
>
>
> ________________________________
>  From: Garot Conklin <garotconklin@stripped>
> To: Garot Conklin <garotconklin@stripped>; "webmaster@stripped"
> <webmaster@stripped>; 'Martin Gainty' <mgainty@stripped>;
> "mysql@stripped" <mysql@stripped>
> Sent: Tuesday, August 21, 2012 4:40 PM
> Subject: Re: Machine Learning
>
> to elaborate:
>
> I would almost rather qualify from the OS perspective what will work in advance, but
> take into consideration any failing condition.  I may be getting a bit too specific at
> this juncture however as I am already apply this logic to issues I see in MY environment
> rather than total ambiguity, which is a best position here.
>
> I am ultimately expecting the learning algorithm to create new and excitingly
> effective ways of stringing commands together to solve issues, this being a notable aspect
> in and of itself, then creating some centralization based on what the machine determines
> as a best solution.
>
> Seems a relative "snapshot of the system" pre and post execution would be favorable
> as well so long as resources are not taxed as a result.  This discussion is leading me to
> believe that this will be a multifaceted db solution without a doubt comprised of multiple
> layers of abstraction ultimately rolling up into a single master cluster for all the
> higher level analytics and internal testing cycles to then roll out the new primary
> protocols, which may themselves take place in a sub-instance of each possible scenario.
>
> I have also though of populating a db or multiple db's with actual text  in the form
> of .pdf's of actual technical manuals... not certain the benefit of the written language
> abstraction there but the thought process extraction might be an interesting end indeed.
>
>
>
> garotconklin@stripped
>
>
> ________________________________
> From: Garot Conklin <garotconklin@stripped>
> To: "webmaster@stripped" <webmaster@stripped>; 'Martin Gainty'
> <mgainty@stripped>; "mysql@stripped" <mysql@stripped>
> Sent: Tuesday, August 21, 2012 4:09 PM
> Subject: Re: Machine Learning
>
> YES!
>
>
> garotconklin@stripped
>
>
> ________________________________
> From: "webmaster@stripped" <webmaster@stripped>
> To: 'Garot Conklin' <garotconklin@stripped>; 'Martin Gainty'
> <mgainty@stripped>; mysql@stripped
> Sent: Tuesday, August 21, 2012 3:37 PM
> Subject: RE: Machine Learning
>
>
> Ah,
>
> Getting clearer and clearer.
>
> So these ‘nodes’ could ‘learn’ and ‘teach’ at the
> same time – right ? For instance, N1 runs a command in ‘domain’ D20
> which it successful – it send information to node N20 that is the authority on
> domain D20 and N20 records it as success; N5 runs a command in domain D20 which goes
> wrong, and sends info to N20 and N20 records failure and sends a correct call to N5. Is
> this what you have in mind ?
>
> There are a lot of variables to be considered, for instance:
> 1)      Command A version 1 can run very happily on operating system version 2, but
> fails in OS version 1
> 2)      Command A version 1 can run happily on OS version 2 on a 2G RAM, but fail on
> OS version 2 on 1G RAM
> 3)      etc
>
> I think the DB design issues will become straightforward once the model is quite
> clear.
>
> Justin
>
> From:Garot Conklin [mailto:garotconklin@stripped]
> Sent: 21 August 2012 17:14
> To: webmaster@stripped; 'Martin Gainty'; mysql@stripped
> Subject: Re: Machine Learning
>
> 1)Refer to it for ‘knowledge’ (for instance, of what the latest
> version of a command is)
>     It would most likely end up being "central" in this sense:
>         A distributed collection of systems; i.e. (possible defined in roles)
>             DB's
>             FE's
>             REPL's
>             Cache's
>         Each DB would have its own collection of remediation's that would then be
> indexed to populate a central db for trending/correlation etc...
>         The "Collective" itself would function as a single conceptual implementation.
> A VIP for example might be associated with a specific role, say Web FE's and remediate
> only/all of them, but only have some relative access to the core db of say the Network
> from a Primary index perspective to make/draw associations/conclusions to issues at hand.
>
>
> 2)Send their ‘knowledge’ (for instance, of the latest command
> versions) to it for storage and distribution to others
>    Rather than "latest command versions" I envision this to be more encapsulated as
> "latest successful invocation of the command string" and it inverse as well (to
> trend/metric-ize the failures thus lending to perpetual optimization).
>
> I like how this is fleshing out... This is helping me to define what I am really
> trying to accomplish. Thanks very much for everyone responding here, this is wonderful,
> please keep this going...
>
> garotconklin@stripped
>
> ________________________________
>
> From:"webmaster@stripped" <webmaster@stripped>
> To: 'Martin Gainty' <mgainty@stripped>; garotconklin@stripped;
> mysql@stripped
> Sent: Tuesday, August 21, 2012 11:19 AM
> Subject: RE: Machine Learning
>
> Hi Garot,
>
>
>
> Ok, the concept is getting clearer, but let’s bring this down to earth a
> little bit more. I love DB design and problem-solving and am quite curious
> about this.
>
>
>
> Is the idea that you have a central computer (not HAL J) somewhere so that
> other computers can:
>
> 1)      Refer to it for ‘knowledge’ (for instance, of what the latest
> version of a command is)
>
> 2)      Send their ‘knowledge’ (for instance, of the latest command
> versions) to it for storage and distribution to others
>
>
>
> If this is the model, then the knowledge base can build up organically over
> time – I think. Or is this too simplistic ?
>
>
>
> Thanks,
>
> Justin
>
>
>
> From: Martin Gainty [mailto:mgainty@stripped]
> Sent: 21 August 2012 00:25
> To: garotconklin@stripped; webmaster@stripped; mysql@stripped
> Subject: RE: Machine Learning
>
>
>
> When I hear 'AI' I always imagine theres a HAL 9001 behind the scenes that
> is running the show constantly admonishing its creator to "take another
> stress pill"
>
> Sounds like a fun project
>
> Keep us apprised,
> Martin Gainty
> ______________________________________________
> Verzicht und Vertraulichkeitanmerkung/Note de déni et de
> confidentialité
>
>
> Diese Nachricht ist vertraulich. Sollten Sie nicht der vorgesehene
> Empfaenger sein, so bitten wir hoeflich um eine Mitteilung. Jede unbefugte
> Weiterleitung oder Fertigung einer Kopie ist unzulaessig. Diese Nachricht
> dient lediglich dem Austausch von Informationen und entfaltet keine
> rechtliche Bindungswirkung. Aufgrund der leichten Manipulierbarkeit von
> E-Mails koennen wir keine Haftung fuer den Inhalt uebernehmen.
>
> Ce message est confidentiel et peut être privilégié. Si vous
> n'êtes pas le
> destinataire prévu, nous te demandons avec bonté que pour satisfaire
> informez l'expéditeur. N'importe quelle diffusion non autorisée ou la
> copie
> de ceci est interdite. Ce message sert à l'information seulement et n'aura
> pas n'importe quel effet légalement obligatoire. Étant donné que
> les email
> peuvent facilement être sujets à la manipulation, nous ne pouvons
> accepter
> aucune responsabilité pour le contenu fourni.
>
>
>
>
>
>   _____
>
> Date: Mon, 20 Aug 2012 13:50:04 -0700
> From: garotconklin@stripped
> Subject: Re: Machine Learning
> To: webmaster@stripped; mgainty@stripped; mysql@stripped
>
> Ya the idea is not anything new, but must be apparently quit difficult or
> not a priority as I have yet to find it already implemented anywhere... Far
> be it from me to not make some attempt here anyway...
>
>
>
> I am creating a fully automated framework from which a distributed
> infrastructure can be maintained.  I have been writing automation
> scripts/code for some time now and the logical progression is to embark on a
> full concept of systems health auto remediation.  I have numerous
> "monitoring" solutions under my control however none that properly (in my
> opinion) implements any real learning algorithms from which to draw even a
> minimalists view of automation.  I like mySQL therefor began thinking about
> creating the aspects (lobes) of the "brain" as a relational database(s).  So
> this is only one facet of what I am trying to do, however leveraging a full
> command set of shell utilities/commands/programs seemed to be a good
> starting point before I get into the "hard" stuff !
>
>
>
> -Garot
>
>
>
> garotconklin@stripped
>
>   _____
>
> From: "webmaster@stripped" <webmaster@stripped>
> To: 'Martin Gainty' <mgainty@stripped>; garotconklin@stripped;
> mysql@stripped
> Sent: Monday, August 20, 2012 3:55 PM
> Subject: RE: Machine Learning
>
>
> Hi Garot,
>
>
>
> You'll have to elaborate some more ... I understand you may want to protect
> the idea as well, so if you can narrow it down to some technical specifics
> then it'll help.
>
>
>
> What is the objective of this system, for instance ?
>
>
>
> Thanks,
>
> Justin
>
>
>
> From: Martin Gainty [mailto:mgainty@stripped]
> Sent: 20 August 2012 19:23
> To: garotconklin@stripped; webmaster@stripped; mysql@stripped
> Subject: RE: Machine Learning
>
>
>
>
>
> From: garotconklin@stripped
> Subject: Re: Machine Learning
> To: webmaster@stripped; mgainty@stripped; mysql@stripped
>
>
> My initial thought was to propagate the db with everything and allow the
> algorithm to then begin to determin trends/patterns
> MG>which trends or patterns will you be modelling?
>
> and begin either an indexing methodology
> MG>which indexes are you considering: Unique index, primary index or foreign
> index?
>
> additional table/db creation process or both to further optimize the calls
> being made
> MG>optimize based on execution time or diskspace allocated, EliminatingFTS
> or some other criteria?
> MG>https://dev.mysql.com/doc/refman/5.5/en/optimization.html
>
> and build in some internal levels of redundancy.
> MG>what about replication
> MG>http://dev.mysql.com/doc/refman/5.5/en/replication.html
>
> I am actually approaching this with some degree of biological conception in
> the multipathing within our own brains however until I have something up and
> running under some substantial load however I may not get a complete
> picture.
>
> Thanks,
>
> Garot
>
>
> Interesting
> Martin
>
>   _____
>
> From: webmaster@stripped <webmaster@stripped>;
> To: 'Garot Conklin' <garotconklin@stripped>; 'Martin Gainty'
> <mgainty@stripped>; <mysql@stripped>;
> Subject: RE: Machine Learning
> Sent: Mon, Aug 20, 2012 7:13:25 AM
>
>
> Hi Garot,
>
> This sounds an interesting idea.
>
> Are you looking to store all known commands and their options or are you
> looking for a 'formula' for calling any unix command ?
>
> The reason for my question is that, at the end of the day, a unix command is
> just a program that is run in the operating system. Each program comes with
> its own options and acceptable inputs. I don't know if there is a rule or
> convention for structuring these commands.
>
> Are you then looking to build a system that 'knows' all commands and 'how
> to' call them ?
>
> Thanks,
> Justin
>
> -----Original Message-----
> From: Garot Conklin [mailto:garotconklin@stripped]
> Sent: 20 August 2012 03:39
> To: Martin Gainty; mysql@stripped
> Subject: Re: Machine Learning
>
> The initial goal is to provide a working framework from which to call all
> UNIX shell command combinations as the underlying storage mechanism for a
> machine learning algorithm.  I would like to build a completely self aware
> instantiation that will maintain itself on all levels... I postulate that
> the first place to start would be in determining a method for maintaining
> all possible remediation combinations including the unknown to eventually be
> learned from and populate new knowledge into the database.  Thank you for
> the reply,
>
> Garot
Thread
Machine LearningGarot Conklin19 Aug
  • RE: Machine LearningMartin Gainty19 Aug
Re: Machine LearningGarot Conklin20 Aug
  • RE: Machine Learningwebmaster20 Aug
Re: Machine LearningGarot Conklin20 Aug
  • RE: Machine LearningMartin Gainty20 Aug
    • Re: Machine LearningGarot Conklin20 Aug
    • RE: Machine Learningwebmaster20 Aug
      • Re: Machine LearningGarot Conklin20 Aug
        • RE: Machine LearningMartin Gainty20 Aug
          • Re: Machine LearningMichael21 Aug
            • RE: Machine LearningMartin Gainty21 Aug
              • Re: Machine LearningGarot Conklin21 Aug
          • RE: Machine Learningwebmaster21 Aug
            • Re: Machine LearningGarot Conklin21 Aug
              • RE: Machine Learningwebmaster21 Aug
                • RE: Machine LearningMartin Gainty21 Aug
                • RE: Machine LearningMartin Gainty21 Aug
                • RE: Machine LearningMartin Gainty21 Aug
                • Re: Machine LearningGarot Conklin21 Aug
                  • Re: Machine LearningGarot Conklin21 Aug
                    • Re: Machine LearningGarot Conklin22 Aug
                      • Re: Machine Learningshawn wilson23 Aug
RE: Machine LearningMartin Gainty23 Aug
  • Re: Machine LearningGarot Conklin23 Aug
    • RE: Machine Learningwebmaster24 Aug
      • RE: Machine LearningMartin Gainty24 Aug
      • Re: Machine LearningGarot Conklin24 Aug