home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cs.utexas.edu!rutgers!andromeda.rutgers.edu!holowcza
- From: holowcza@andromeda.rutgers.edu (Richard D Holowczak)
- Newsgroups: comp.databases
- Subject: Re: References wanted on Distributed Databases
- Message-ID: <Dec.17.22.49.53.1992.22812@andromeda.rutgers.edu>
- Date: 18 Dec 92 03:49:56 GMT
- References: <1992Dec10.170708.5610@cs.brown.edu> <kitchel.724014868@manta> <1992Dec11.080514.17132@cs.ruu.nl> <1992Dec11.114955.1@bigez> <Dec.15.16.02.57.1992.27495@andromeda.rutgers.edu> <1992Dec16.175436.1@bigez>
- Organization: Rutgers Univ., New Brunswick, N.J.
- Lines: 63
-
-
- dmmatt@bigez (Mike Mattix) writes:
-
- >In article <Dec.15.16.02.57.1992.27495@andromeda.rutgers.edu>, holowcza@andromeda.rutgers.edu (Richard D Holowczak) writes:
- >>
- >> I think there is a big difference between distributed partitioned data and
- >> distributed replicated data. I would be interested to know how Rdb (and
- >> others) handle site failures, communication failures and recovery.
- >> Any comments?
- >>
- >>
- >>
- >> Rich Holowczak
- >> Rutgers University
- >> holowcza@andromeda.rutgers.edu
-
- > Since I brought up Rdb (I just get sick of hearing about the NEW ADVANCED
- >features of Oracle/Ingres), I probably should take a stab at this. The
- >application we did originally had distributed partitioned data (multiple
- >table of different data on different network nodes). The application
- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
- O.K. This can be (and is being) done by many vendors. As to who did it
- first, I have no idea. The problem with performance you state is
- remedied by replicating data. The goal (via, say one-copy token approach)
- is to dynamically move the data close to the people who need it while
- still allowing access from all others who might. This is a tricky bit
- of business which I don't think Rdb or others is up to.
-
- The new Sybase release is supposed to handle this although not very
- gracefully from a theoretcal standpoint.
-
-
- >worked fine we ended
- >up consolidating the application on one node for performance reasons. I don't
- >believe many people will argue that distributed applications tend to
- >perform worse
- >than applications where all the data resides on one node.
-
- See above.
-
- >The failure of a node is
- >handled by the two phase commit function during a transaction.
- >The failure of a node
- >is handled by error trapping after that, ie the application will not be able to
- >regards,
-
- >--
- >Mike Mattix
- >Agricultural Group of Monsanto
- >P.O. Box 174
- >Luling, LA 70070
- >INTERNET Address: dmmatt@bigez.monsanto.com
-
- ---
- [1mRich Holowczak[m holowcza@andromeda.rutgers.edu
- Rutgers University
- Ph.D. Computers and Information Systems Program
- "Availability of Distributed Databases During Network Partitioning"
- +---------------------------------------------------+---------------+
- | The .sig, the whole .sig and nothing but the .sig | Ask about the |
- | <<< This space for rent >>> | car alarm FAQ |
- +---------------------------------------------------+---------------+
-