mesos-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Heller, Chris" <chel...@akamai.com>
Subject Re: [DISCUSS] Renaming Mesos Slave
Date Wed, 03 Jun 2015 18:24:17 GMT
In all truth, I think the reason this idea even has legs is that the
master/slave labels are not really accurate for the two mesos servers
based on their actual roles in a mesos cluster.

That said, I¹m not certain the roles can each be easily distilled down
into a single catch-all word.

"There are only two hard problems in Computer Science:
   cache invalidation and naming things."

   -- Phil Karlton


-Chris


On 6/3/15, 3:17 PM, "CCAAT" <ccaat@tampabay.rr.com> wrote:

>On 06/03/2015 09:27 AM, Brian Hicks wrote:
>> 1. We¹ve been aliasing everything to ³leader/follower² in the projects
>> I¹m a part of. I think it gets rid of the problematic language while
>> still being descriptive of the architectural model. ³agent² makes sense
>> to me, too (since those processes do have agency in some situations) but
>> I would call it a second choice, personally.
>
>With all due respect, I think my concerns are being ignored, so I going
>to attempt to shed some light on them, again.
>
>Backgound: Although most of my work is in software, Computer scientist,
>I also hold degrees in Electrical Engineering and PetroChemical
>Engineering. I am "a hardware guy" who writes code and implemnents
>technology long before it becomes an industry standard. As an older and
>definitely wiser engineer folks are not recognizing the problems that
>lead to 'shear idiocy' with some perceived benefit.
>
>IN computers we have busses. This is always a buss-master and slaves
>that participate in Buss arbitration as slaves are call
>'slave-processor. There are many analogies where you are going to leave
>hardware designers scratching their heads and laughing, loudly.
>
>Master-Slave is as much a mathematical term among its myriad of
>connotations. It means that master (A) 100% controls slave (B) in a
>fully deterministic fashion. It means slave (B) is 100% subservient to
>any future requirement yet to be defined or implemented by Master
>(A).It's a mathematical thing.
>
>I do not believe there is another word-pair in the English (globish)
>language that remotely displays this level of control nor instantaneous
>comprehension. Technology is built upon jargon; you mess with that you
>risk obscurity. The history of parallel process is a grave-yard of
>idiodic terminologies.
>
>Ah *CONTROL THEORY*.  Master-Slave is as fundamental as arithmetic. In
>the industrial world there are billion if not trillions of master-slave
>relationships between hardware, systems and software. It's 100%
>deterministic (non-fuzzy unless you count 'dont-care states') for hard,
>practical reasons. Often something that acts autonomous, is changed and
>brought into "slave mode" to ensure correctness and determinism. A
>common example, If you have several process controllers in a series, the
>controller at the start of the arrow is always the 'master' and the
>controller at the point of the arrow is the slave. Instantly understood
>and universally accepted.
>
>It is a convention in hardware and software that is instantaneously
>understood. You change that nomenclature and folks are just going to
>scratch their head, stick up their hand and say isn't this just a
>master-slave relationship? Once you affirm that, then their second
>question is the totally predictable, so why not call it master-slave
>like everything else does? I could write textbooks on the engineering
>usage of 'master-slave' and never have I as an engineer with a
>Professional Certification have I never heard of this complaint about
>such a prcise meaning in terms. In engineering and science, english is
>the best and universal language, but it is mostly a struggle to approach
>such accuracy as the terms 'master-slave'. Please do not 'fuzzify' this
>well understood and universally accepted jargon. Surely
>you and others can find a better pathway for social improvement?
>
>
>Hardware-->Software-->applications.
>-----------------------------------
>
>I know this is a 'software-weenie' intensive group here developing and
>testing mesos; surely I fall into that category too. Hell, I build
>everything up form 100% sources as a gentoo aficionado. But, clusters
>and clouds will make their biggest splash in the 'hardware world', once
>ubiquitous computations and general software can be migrated to a
>cluster. Mesos is a candidate, but sticking with established industry
>jargon, is a very good idea, imo.
>
>Process Controls (the software, hardware and logic used to control
>pumps) by some estimates are responsible for more that 40% of all energy
>consumption on the globe. There is a gargantuan need to apply
>clustering, particularly ubiquitous clusters, so that faster and better
>software can have more features and support things like R, Scala, and
>Haskell. I'm already on contract to deliver a mesos cluster, to a
>company full of technicians to use for 'sensor fusion' and 'video
>analytics' Most technicians struggle with windows and macs yet every
>thing I've set up for them before on linux or bsd is wonderfully
>accepted and operated by *TECHNICIANS*. Please go spend some time in the
>field and deploy some state of the art software and train technicians to
>use, to 'master it' and how to fix it in the middle of the night so you
>can get some sleep (thank GOD for redundancy and hot spares). Cameras
>and sensors are 'slaved' to processes and controllers and software,
>routinely.
>
>I also am working with some computational chemistry folks (at a
>university) for customizing a cluster with very specific and bontiful
>embedded systems (FPGAs). Sorry, but the term master-slave is of
>paramount importance systemically in that project too.
>
>
>
>As a public speaker, as a consultant, as the defacto resident expert on
>matters of software-hardware interaction where ever I go, much to my
>chagrin, I can tell you unequivacolly I have never ever heard anybody
>complain about the term, master or slave or master-slave, in the many
>heterogeneous environments over the deecades from the deserts, to the
>arctic and places that do not formally exist. *NEVER*. But I will
>concede I avoid idiot psycologists and social engineering to those that
>like to *waste the time* of others. Me, I have to 'getter done' and
>master slave is not optional, its a fundamental mathematical
>relationship, since long before the 'Greek masters' and mathematicians.
>
>
>
>
>> 2. mesos-{whatever-we-decide}
>> 3. I don¹t think ³master² is as problematic as ³slave², but it does have
>> some of the same connotations. If we¹re going to take the time to
>> rename, we should rename both. But if it turns out to be infeasible from
>> a project management standpoint, ³master² will probably be fine.
>> 4. same as others have said: start off with documentation changes,
>> follow through with code changes. Deprecate the old binary names over
>> several versions, to avoid a cliff.
>
>Masters only have slaves. A master is 100% in control or some form of
>autonomy exists. So if you have a master, you have slaves. What level of
>math did you finish in school? Did you go to a good university or night
>school? You need to brush up on math, for this audience, imho.
>
>I previously suggested that this naming scheme be set up so as to be
>'user definable'. You know may 'preprocessor directives' or such. This
>would ease the transition across a few releases of mesos and allow for
>folks like myself to use industry standard jargon, i.e. to use correct
>terms with instantaneous understanding, in a fully deterministic manner.
>Where is the discussion on that propose solution? This solution
>of user definability is a win-win for all users of mesos. It also set
>the stage in the future for nodes that have some degree of well define
>autonomy or even experimentation with 'self-modifying-codes' among the
>slave populations. It even sets the state for node migration amoung
>differnent cluster solutions in data centers.
>
>So if you are fully determined to change the name, why not add some
>flexibiilty for the evolution of mesos, but allowing these and other
>terms to be 'user definable'?
>
>
>
>> For what it¹s worth, I think it¹s good that the Mesos project
>> understands that the words we use have different meanings to different
>> people. While I am not ³offended² by the current terminology I
>> understand that master/slave has a strong connotation towards ongoing
>> civil and human rights abuses, and I would seek to distance the project
>> from that. Continuing to make these words part of our everyday lexicon
>> normalizes oppressive structures, and sends a strong signal to people
>> outside the ³blessed² group.
>
>
>BLAH BLAH BLAH. Really? You sound like a pedantic defining a social evil
>that  fills  you full of guilt;but yet to instantiate for the rest of
>society. Can you name one actual person that is offended by the terms
>master-slave that will actually peruse the sources? Note: sun-baked
>Californians with little productive requirements at their job, do not
>count. Boy, I guess the drought is really tougher than the rest of us
>realized?
>
>
>> Mesos is an industry-leading technology,  and we have a responsibility
>>to build an inclusive
>> and friendly  community. This is not only for the good of the product,
>>but for
>> the  good of the industry and all the people whose lives are effected
>>by it.
>
>Not if idiotic ideas like this dominate. I suggest you weigh this
>perceived benefit against the technical costs and the shear fact that if
>these 'workers' are 100% control, they are slaves in ALL common industry
>jargon. No wonder EEs laugh at CS folks. Just so you know
>there are EE that are building and operating computers that run on
>on waves (light). You might also directly comment on the idea of making
>these terms 'user definable'. Or is this just some form of tantrum?
>
>
>> So, strong +1.
>> Brian Hicks
>
>So a realistic and practical -1
>Sincerely
>James
>

Mime
View raw message