mao@eden (Mike Olson) (09/22/90)
In <1124@beguine.UUCP>, robinson@durham.med.unc.edu (Gerard A. Robinson) writes > How about some fun :-) A while back, in this group, Dr. Stonebraker made > the statement that database servers will be so plentiful that they'll be > a-dime-a-dozen. I disagree, I think that *GOOD* db servers (i.e. rock-solid > reliability with fast response time) will be so rare, that THEY'LL decide > the marketplace. (Actually, most users will compromise on the speed, if > that's what's needed to get the rock-solidity.) i'll bite. you should probably discount my opinion, since i work for dr. stonebraker. i believe that the evolution of servers in the industry toward ansi SQL and generally identical functionality among vendors reinforces stonebraker's claim. most commercial database systems *are* rock-solid, in terms of commits succeeding and rollbacks really rolling back. further, although most vendors are going to spit blood at this statement, they're all pretty much the same in terms of speed, if you normalize for the hardware they run on. in what sense is any commercial database system more solid than any other? so why do people choose one vendor over another? because the front-end tools are better. i think the acquisition of ingres corp. by ask is significant. ask builds hot front-end tools for the ingres backend. ingres was having financial problems selling its database system because database systems aren't what attract users. ask, which ignores the database system in favor of selling tools that use it, made heaps of cash. companies like metaphor, who develop tools that run on lots of different rdbms platforms, are in much better shape to survive than are companies who sell a single, specialized product. this is pretty straight darwinism -- diversity is a good thing, and if you eat lots of things low on the food chain, you can survive the extinction of any one of them. rdbms vendors are low on the food chain. as commercial database systems continue to become indistinguishable from one another, the only thing that users will care about is the tools they can get. those tools will, in general, run on all commercial database systems. this is an exact parallel of the open systems revolution happening among hardware and os vendors these days. mike olson postgres research group uc berkeley mao@postgres.berkeley.edu
robinson@durham.med.unc.edu (Gerard A. Robinson) (09/22/90)
My original question was: # How about some fun :-) A while back, in this group, Dr. Stonebraker made # the statement that database servers will be so plentiful that they'll be # a-dime-a-dozen. I disagree, I think that *GOOD* db servers (i.e. rock-solid # reliability with fast response time) will be so rare, that THEY'LL decide # the marketplace. In article <28051@pasteur.Berkeley.EDU> mao@postgres.Berkeley.EDU (Mike Olson) writes: >i'll bite. you should probably discount my opinion, since i work for >dr. stonebraker. > >i believe that the evolution of servers in the industry toward ansi SQL >and generally identical functionality among vendors reinforces stonebraker's >claim. most commercial database systems *are* rock-solid, in terms of >commits succeeding and rollbacks really rolling back. ( some deleted text >... ) in what sense is any commercial database system more solid than >any other? In the sense that other things don't happen. An example from my experience with INGRES 5.0 (not server technology, but still illustrative) is that a btree structure could somehow get "a tiny bit corrupted" so that on a SPARC RISC system which is sensitive to data alignment boundaries, the OS would panic with a "memory alignment" error and crash the system. The problem could be fixed by finding the errant table and re-modifying it (the re-modify would usually succeed although sometimes re-crash the system) although we never found or received a remedy (in favor of V.6.3). No data was lost, but time and energy certainly were. Its these little *extras* that the users don't like, and that *all* db servers (from reading this list) have in some measure or another. So 'rock-solidity' includes (the absence of) these. > >so why do people choose one vendor over another? because the front-end >tools are better. Indeed. This was a large part of our decision to stick with INGRES four years ago, when a new director (and thus new and expanded development push) arrived. > >i think the acquisition of ingres corp. by ask is significant. ask builds >hot front-end tools for the ingres backend. ingres was having financial >problems selling its database system because database systems aren't what >attract users. ask, which ignores the database system in favor of selling >tools that use it, made heaps of cash. > > (deleted paragraph) > >as commercial database systems continue to become indistinguishable from >one another, the only thing that users will care about is the tools they >can get. those tools will, in general, run on all commercial database >systems. this is an exact parallel of the open systems revolution happening >among hardware and os vendors these days. I don't disagree. In fact, this is almost my starting point. OSs are even lower on the food chain than rdbms systems and two is still one too many for most open system proponents. When does INGRES, for example, start producing tools for Oracle and Sybase servers? Then, when do they make the decision to stop, for financial reasons, the development work on their own server, in favor of other ones (if at all, its likely that their server could become fairly ubiquitous)? Gerard Robinson, UNC-CH School of Medicine, Office of Information Systems