sachar@hou2d.UUCP (04/03/87)
Ignoring the user friendly and other issues, network model bases
databases should have better performance than relational model
based databases due to presence of links. Is anyone aware of
any study or published paper where the performance issues of
these models are discussed and some results (numbers) are reported.
I will summarize the responses if anyone else is interested.
Thanks in advance for any information.
Harvi Sachar
!ihnp4!speedy!hselwell@osu-eddie.UUCP (04/04/87)
In article <1361@hou2d.UUCP> sachar@hou2d.UUCP (H.SACHAR) writes: >Ignoring the user friendly and other issues, network model bases >databases should have better performance than relational model >based databases due to presence of links. Is anyone aware of >any study or published paper where the performance issues of >these models are discussed and some results (numbers) are reported. >I will summarize the responses if anyone else is interested. > >Thanks in advance for any information. > > Harvi Sachar > !ihnp4!speedy!hs I haven't done much looking around in the literature, so this isn't really an answer, but I can offer some empirical results from having implemented both relational and network databases. Based on my experience so far, the network model is much faster and more flexible for a given level of (subjective) implementation complexity. I must admit that these databases have been primarily retrieval-oriented, so I was able to play some tricks at the expense of update speed, but this seemed to affect both models similarly. I have almost entirely given up on the relational model for real-world applications, simply because the network model gives better performance for a comparable investment of effort. It should not be forgotten, however, that many applications can be handled by off-the-shelf software, which is predominantly relational. If it can be done, this is often the simplest approach, which is why things like UNIFY and Informix are successful. When it comes down to raw performance, though, they don't quite cut it. That's the price you pay for the time you save in developing it. I'd be interested in hearing about what references you come up with. -=- "The greatest warriors are Clayton Elwell the ones who fight for peace." Elwell@Ohio-State.ARPA --Holly Near ...!cbosgd!osu-eddie!elwell
garyp@cognos.uucp (Gary Puckering) (04/07/87)
In article <1361@hou2d.UUCP> sachar@hou2d.UUCP writes: >Ignoring the user friendly and other issues, network model bases >databases should have better performance than relational model >based databases due to presence of links. This is a commonly-held fallacy. Linkage paths represent an implementation option, just as B-tree, hash and clustered indexes are options that speed up certain queries. There is no theoretical reason why a network dbms should perform better than a relational dbms. Only practical reasons. Most of the systems I've measured tend to perform somewhat below their network counterparts because: 1) They do a lot more (there's more code, etc.) 2) They provide greater data independance (datatype conversions) 3) They provide more sophisticated locking and transaction management (degree 3 consistency, as opposed to degree 2) 4) They provide varying-length datatypes and null-values, which means more overhead in terms of a record management scheme 5) They are immature products The last point is perhaps the most important to bear in mind. Commercial relational systems are still immature in comparison to most network and hierarchical dbms's. Ask someone who was involved when IBM's IMS was introduced. Comparing DB2 performance to IMS performance is like comparing an 8-year old runner to a 20-year old runner. In a few years I thin the performance debate will fade away. -- Gary Puckering 3755 Riverside Dr. Cognos Incorporated Ottawa, Ontario decvax!utzoo!dciem! (613) 738-1440 CANADA K1G 3N3 nrcaer!cognos!garyp