[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: GEU-algorithms (was Ozone object identity)
David Li wrote:
>
> Using a hash value to annote a global object does satisfy the GEU
> (globall and eternall uniquess) properties of an identity. We also agree
> that a URL scheme satisfy that as well.
>
> However, there is a problem with hash value is how to recover enough
> information from it to reconstruct the object.
>
> Take the discountinuous mobile database example. I get objects from
> several sources, keep the locally, disconnect from the server, modify
> the data, and later on sync to the server.
>
> In this implementation (as well as the ClinetCacheDatabase already
> proposed for Ozone), the implementation must be able to retrive the
> object later on from the identity.
>
> Using a hash value as identity will likely to result the following
> implementation:
>
> 1. A hostname, port, database name, time, and a random value are hashed
> to generated the id.
The port here covers the idea of having multiple ObjectBases running on
the same machine unless you create a listener like Oracle uses that lets
you run multiple DBs(known as SIDs) using one listener and thus a known
port. The uniqueness is the port plus the SID in this case which is in
the JDBC URL. Multiple instances of ozone running in the same VM could
possible save resources. Of course each instance would have to have it's
own classloader.
>
> 2. The id is sent down to the client side
>
> 3. The client side cache must associate at least hostname, port and
> database name with an object identity in order to recover the objects
> later on while connecting back to the real database.
>
> Why we destroy the perfect information in the beginning just to have
> client side having to do more work. Won't encoding the necesary
> information in the identity be better?
>
> CORBA (www.omg.org) has a similar scheme in producing the ID for a
> distributed object. Full information about the object (host, port, and
> id) are encoded in an object id. A client side app can retrive the
> object using just the id.
>
> I think we are looking for the following properties in a identity.
>
> 1. Globally and Eternally Uniqueness
>
> 2. Locatablility (be able to locate a where about and access the object
> from the identity)
>
> I think we have agree on the first one. Now, let's see if the second is
> an important property in our search for perfect object identity
> representation.
>
> David
>
> Ola Berg wrote:
> >
> > There are a couple of algorithms that almost ensures
> > global and eternal uniqueness. By "almost" I mean that
> > the chances for collisions are sooooooo minimal.
> >
> > They all built on secure hashing. You feed the algorithm with
> > whatever uniqueness information you have (millis since the epoch,
> > hostname, data from the object, object counter in the VM blablabla.
> > The hash it produces will probably be unique.
> >
> > How? The hashing algorithm shuffles the input totally (converts
> > the bits into noise in an orderly fashion). Since the input is
> > an ordered data combination, and there are billions of noisy
> > combinations for each ordered combination, the hash will probably
> > be unique.
> >
> > The Secure Hashing Algorithm (SHA) has this property, and it produces
> > 20 bytes. A byte[] of size 20 will for all practical purposes be
> > globally and eternally unique.
> >
> > The idea (and I am sure that Falko didn't have this in mind from the
> > beginning) would be that each and every instantiated Ozone object will
> > have a GEUID. Exactly how this is implemented should be hidden, but
> > identity should definitely be both in time and space.
I'm sure glad someone knows about these GEUIDs as otherwise my
discussions would be kind of suspect and this is certainly out of my
comfort zone. Did they change the terminology to add the E to put in the
time consideration?
> >
> > I am thinking about the ability to move ozone objects between
> > several ozone servers (for load balancing maybe, or for distributed
> > computing, peer-to-peer capability of ozone where the objects float
> > around the network until it finds computing space, or...). The point
> > is that object identity is a good thing, and to not mess up things
> > in the future, let's do it GEU.
> >
> > Furthermore, a standard URL scheme is also a good idea. But let's not
> > confuse this with GEU. URL's are bad candidates for GEUness. But if
> > you have a GEUID, it is easy to implement a URL scheme on top of it.
The Handle object in EJB has the property that allows lookup etc very
much like the CORBA OID. BTW, the EJB method is boolean
isIdentical(EJBObject) which is similar to the discussed boolean
isSame(OzoneRemote?). So the URL and port and ID must be saved as
mentioned by David.
Eric