Archived

This forum has been archived. Please start a new discussion on GitHub.

Connections #2: IceGrid

Hi !

Michi's "To Slice or Not to Slice" is an excelent and important article and I think the reference to it should be on the main ZeroC page near the reference to "Ice vs Corba".

My main comments concern the article about IceGrid and they are in the attachment. Briefly, from my point of view the proposed service has remote relation to the Grid computing and the name 'IceGrid' is misleading.

Hope my comments can be useful although, as I already mentioned in other post, different people have different opinions on what is Grid.

Cheers, Nikolai

Comments

  • The IceGrid clearly is a grid.

    Foster and Kesselman the gentlemen behind "Globus" made this definition back in 1998 before it became a buzz word or a funding concept!

    "A computational grid is a hardware and software infrastructure that provides dependable, consistent, pervasive, and inexpensive access to high-end computational capabilities"


    Since then the idea has grown beyond computational grids to include data grids, commodity grids, database grids .... This is the problem with Globus, its attempt to bring this all together has made it a mess, quite similar to CORBA in a way.

    Based on your definition then, Sun Microsystems should not be calling its N1 Grid Engine a grid, then why does the Global Grid Forum consider it so? I don't have a problem with the Sun calling its product a grid even if I do consider it identical to your "lightweight grid" and nothing more than a job scheduler.

    The point is that a grid must exploit geographically distributed resources regardless of resource type and seperation, that includes a homogeneous cluster of PCs in a small room to heterogeneous super computers spread across the world. Grids are not the sole domain of those dictating a narrow definition that must include security and resource location or other specifications. These are features that may or may not be included in a grid environment and define its functionality not whether or not its a grid.

    Consider NetSolve, Grail, The BioGrid, The Swiss BioGrid, GridForge, RedGrid, GridBus ..... some of these don't even match your security/resource location prerequisite yet have been accepted so by the greater scientific community in a plethora of papers.

    I think stateing that IceGrid is not a grid, is similiar to saying that the only way to do particle physics is to use a large hadron collider. It would be prudent to realize that the Connections one page article is a product teaser, not a scientific article and to reserve judgement until the source is released and its functionality can be tested.
  • bernard
    bernard Jupiter, FL
    Thanks for your feedback! We want to release IceGrid soon, so we need to limit the scope of the first IceGrid version. Over time we'll add more features to satisfy even more use-cases.

    Yes, IceGrid will be just for Ice servers and Ice clients: our primary focus are Ice applications, this should not come as a surprise. If you want to manage a non-Ice "job", you can write an Ice server that forks and monitors this job -- you probably have such a server in MG.

    Fig 2 in the Connections article shows replication, not resource location. IceGrid will provide some resource location features/APIs, but we haven't decided on them yet.

    Regarding security, you're right, a secure deployment over an untrusted network often needs features (authorization, auditing) that are not provided by Ice or IceGrid. We will address this limitation in a subsequent release.

    Cheers,
    Bernard
  • The point is that a grid must exploit geographically distributed resources regardless of resource type and seperation, that includes a homogeneous cluster of PCs in a small room to heterogeneous super computers spread across the world.

    Absolutely agree with this general statement.

    Before we can make next step we should ask "what's a difference between
    Grid and any distributed program ?"
    Is CORBA-based "hello, world" server a Grid ? If no, why ?
    Is 'rsh' a Grid ? If no, why ?
    Is LSF a Grid ? If no, why ?
    Is 'Condor' a Grid ? If no, why ?
    Is 'Condor-G/C' a Grid ? If yes, why ?
    (Related question: what kind of changes the Condor team made to move from
    Condor to Condor-G and why ?)

    The most important feature which differs Grid computing from other
    similar kinds of distributed computing is that the participating parties,
    i.e. user's applications and resources, are geographically dispersed and
    under different administrative control.
    The consequence of this feature is that the user does not know in advance
    about all available in Grid resources. As well as resources have no ideas
    about user's applications. And the next logical consequence, now for
    architects of the Grid systems, that the match-maker
    (application requirements <-> resource properties) component as well as
    high level of security are mandatory. The architect can think about
    different possible designs of match-maker (for example, "classical"
    Grid vs P2P Grid) or about the needed level of security (for example,
    practically all problem domains require authentication but not all
    require encryption).

    To build general purpose Grid is a big, very big challenge. That's why
    all current Grid infrastructures implement only a subset of what is needed.
    There are a lot of research projects in this area.
    And this is not a problem if some system is called "grid" as soon as
    it clearly specifies what kind of grid is really implemented.

    The story is different for Ice.
    Ice is an excellent product itself, but also, from my point of view,
    is the best (!) choice as a middleware platform for the Grid Computing.
    The Grid Computing is an extremely wide area. I believe that sooner or
    later there will be numerous Grid infrastructures based on Ice and the
    Ice services. Many of such systems will include deployment and monitoring
    services where the discussed Ice service will help a lot.
    But I also think that the name 'IceGrid' in the context of such systems
    will be misleading for users.
    I also think that the name 'IceGrid' for the proposed service can create
    a problem for the ZeroC developers themselves if in the future they will
    add new Grid-related services.

    Cheers, Nikolai