morse@quark.mpr.ca (Daryl Morse) (02/12/91)
In all of the reading on OSI (both standards and otherwise) that I have done, I have noticed what seems to be an inconsistency in the conceptual model that forms the basis for the OSI. The seeming "inconsistency" is as follows: In both name, and in the actual protocols that are in the stack, OSI is pretty clearly aimed at the connection of systems. We can see this by virtue of the types of end to end communications that have been standardized. All are aimed at system to system communication. However, if you look at the structure of the application layer, the picture seems to change. Indeed, the name of the upper-most layer seems to verify that the OSI is also aimed at the connection of applications. This is reinforced by the preliminary work that I have reviewed on the ODP (Open Distributed Processing) Reference Model. This also seems to be clearly aimed providing transparent distribution at the application level. Why, then, is there no light-weight mechanism for applications to communicate with other applications that might happen to be located on the same system? If I don't want my application to know where it is, or where the applications that it communicates with are, shouldn't that be addressed (no pun intended) by the stack? Surely, I shouldn't be reasonably expected to use X.25, or FDDI, for IPC. (Something related to this is the lack of a decent performing name service. X.500 is clearly not a great name service for an RPC.) Am I all wet here, and it really isn't the intention of the OSI that application entities might reside on the same system? Perhaps I'm just impatient, and these issues will be addressed by the ODP work? If you have any thoughts on these matters, I would be most interested in hearing them. Thanks. -- Daryl Morse | Voice : (604) 293-5476 MPR Teltech Ltd. | Fax : (604) 293-5787 8999 Nelson Way, Burnaby, BC | E-Mail: morse@quark.mpr.ca Canada, V5A 4B5 | quark.mpr.ca!morse@uunet.uu.net
mdobbins@dvnspc1.Dev.Unisys.COM (Michael Dobbins) (02/13/91)
In article <MORSE.91Feb12152547@quark.mpr.ca> morse@quark.mpr.ca (Daryl Morse) writes: >[stuff deleted] >Why, then, is there no light-weight mechanism for applications to >communicate with other applications that might happen to be located on >the same system? If I don't want my application to know where it is, >or where the applications that it communicates with are, shouldn't >that be addressed (no pun intended) by the stack? Surely, I shouldn't >be reasonably expected to use X.25, or FDDI, for IPC. (Something >related to this is the lack of a decent performing name service. X.500 >is clearly not a great name service for an RPC.) > >Am I all wet here, and it really isn't the intention of the OSI that >application entities might reside on the same system? Perhaps I'm just >impatient, and these issues will be addressed by the ODP work? My understanding of the intention of OSI (at least a couple of years ago) is to standardize Intersystem communication. Any intrasystem issues were considered private to that system and not to be defined by OSI. This includes layer boundry interfaces. It is not inconsistent for a system to provide a local IPC mechanism which appears the same as the remote IPC mechanism to an application. IMHO, that is the right thing to do. The local mechanisms have not been within the scope of OSI definition. On the other hand, other groups have been working on standardizing local system interfaces and API's, ex. TLI for unix. This work may result in the 'defacto' standards for local 'light weight' mechanisms that you are looking for.
eskovgaa@CUE.BC.CA (Erik Skovgaard) (02/14/91)
OSI is intended to be used between processes on different systems.
Whatever takes place on a single system is outside the scope of OSI, but
as you point out, this will be taken up in the ODP framework.
Other organizations such as X/Open, IEEE and the API Association have
decided to standardize APIs for internal use on a computer system. As
it turns out, many of these APIs reflect the Abstract Services of
several ASEs.
....Erik.
---------------------------------
Erik Skovgaard
PSC (Pacific) Inc
Vancouver, B.C.kerry@batserver.cs.uq.oz.au (Kerry Raymond) (02/18/91)
In article <MORSE.91Feb12152547@quark.mpr.ca> morse@quark.mpr.ca (Daryl Morse) writes: >OSI is pretty clearly aimed at the connection of systems. >This is reinforced by >the preliminary work that I have reviewed on the ODP (Open Distributed >Processing) Reference Model. This also seems to be clearly aimed >providing transparent distribution at the application level. >Why, then, is there no light-weight mechanism for applications to >communicate with other applications that might happen to be located on >the same system? One of the three main aims of ODP is to provide distribution transparency which should blur the distinction between local and remote. To that end, ODP wants to see communication between co-located and non-co-located applications done via the same interface. Whether the implementation of those interfaces is smart enough to realize that some light-weight mechanism can be used in certain circumstances depends on those implementing the lower-level protocols. I suspect that the problem is that the lightest-weight mechanism is usually the local IPC mechanism. However, the implementer wants to write something very portable which can run on many machines and hence isn't interested in including local IPC calls. >Perhaps I'm just impatient, and these issues will be addressed by the ODP work? I doubt that ODP will address the light-weight mechanism issue. I believe that the ODP group would regard that as a problem for lower layers of the OSI stack. =============================================================================== Dr Kerry Raymond, R & D Engineer kerry@citrus.citr.uq.oz.au Centre for Information Technology Research Phone: +61 7 365-4321 University of Queensland 4072 Australia Fax: +61 7 365-4399