RE: Current Informix vs. Oracle Data

Take a look at this slightly dated paper http://www.oracle.com/st/o8collateral/html/xoustwp5.html --- Sathish S -----Original Message----- From: Donald Johnson [mailto:rp2360@email.sps.mot.com] Sent: Wednesday, February 17, 1999 11:10 AM To: informix-list@iiug.org Subject: Re: Current Informix vs. Oracle Data Just a few comments ... Informix as a company is MUCH easier to work with. While the base license costs for the two are the same ( or very close ) Informix bundles parititioning which Oracle DOES NOT and wont give a date/release as to when they will ( my own personal soapbox item ). 3rd party tools are more prevelant to support Oracle but companies like Platinum are beginning to see the light after the troubles of the last couple of years and supplying tools to make our jobs as DBA's easier. Based on my last working with support of the products, you can expect Informix to be better - but still not as good as we would all like for what we pay. -=Eclypse=- wrote: > > I'm looking for someone who can give some good input in the Informix vs. > Oracle arena. Management is all up on Oracle because "everyone knows how > to use Oracle and it's popular and try to find an Informix DBA - there > aren't any, but if you look up Oracle DBAs, they can be found easily." > I really can't compare the two because I have never seen Oracle, never > used Oracle, and haven't seen any information on recent versions of the > two RDBMS's that compares the two. My thought seems to be that it's like > the VHS vs. Betamax deal - Beta (informix) was a much better format, but > VHS (oracle) won anyway because more people used it. I like Informix, it > has never given me any problems, it does everything we need, but the new > app they want to develop is something they don't want to have to look > too hard to find a good DBA (or two) for and they think Oracle is the way > to go. The Oracle sales nazis have brainwashed them pretty good and there's > nothing I can say to sway their opinions. Now the Oracle people have offered > to swap out Informix for us and replace it with Oracle. My thing is that > everything they've mentioned that Oracle is so great at, Informix will do. > Their thing is, where do we find people who know Informix? I would just > like somewhere to look for information to hand them that will get them > to consider something else. Any help/information is appreciated... > > -- > > Jon<*> -- Tony Johnson Email : rp2360@email.sps.mot.com Senior Database Administrator Voice : ( 602 ) 655 - 4617 Motorola SPS Information Technology Pager : ( 888 ) 772 - 5229 2200 W. Broadway Rd. MD M555 Fax : ( 602 ) 655 - 3569 Mesa, AZ 85202 ---------------------------------------------------------------------------- ------- We Are One. One Team. One Voice. One Objective : To Be No. 1 With Our Customers. No RAS. DWYSYWD. LOPOYPDNNMAEOMP. Top Box. WDIRTFTWYCDIA. No Necktie Damage. Murphy's Data Constant: Data will be damaged in direct proportion to its value

"Sadagopan, Sathish, CFCTR" wrote: > > Take a look at this slightly dated paper > http://www.oracle.com/st/o8collateral/html/xoustwp5.html > OK Here goes... A counter on Oracle's dated white paper. This refers to Figure 1: Stored procedures - supported in 8.2, and 8.3 ( 8.21.UD3 now shipping ) Procedural Language - do I need this? What about using whatever language you wanted? OLTP Support - Uh, what does that mean? Of course you can use 8.x for OLTP. It's OK, go ahead, you won't hurt it. I hear a couple of sites are actually using it for OLTP. Warehouse Support - Define please. Any Informix engine could conceivable be used, however only 8.x is parallel. It is highly doubtful that Oracle is anything close to what Informix has, simply because of the struggle that ORCL went through just to get SMP and MPP to work, if at all. Replication - In many instances totally inappropriate in a data warehouse environment. Feeds from the DW can go to any other Informix engine, but Replication may not be appropriate even if the option is available. Parallel ins/upd/del - Uh, how long did you want to wait on Oracle? Their HP loads take eons longer than Informix. Extensibility - Shucks, ya caught me! 8.x doesn't have it now, but maybe you need to ask why you think you need it. When engines merge in Informix, I hear that 8.x will not be a part of the merger, simply because of the divergent architecture. But having 7.x and 9.x merge will be nice. At any rate the IFMX architecture is still light years ahead of Oracle. You might even want to ask why you'd do things so badly with Oracle when Informix is so much more flexible. <...> > Tony Johnson Email : > rp2360@email.sps.mot.com > Senior Database Administrator Voice : ( 602 ) 655 - 4617 > Motorola SPS Information Technology Pager : ( 888 ) 772 - 5229 > 2200 W. Broadway Rd. MD M555 Fax : ( 602 ) 655 - 3569 > Mesa, AZ 85202 > ---------------------------------------------------------------------------- > ------- > We Are One. One Team. One Voice. One Objective : To Be No. 1 With Our > Customers. > No RAS. DWYSYWD. LOPOYPDNNMAEOMP. Top Box. WDIRTFTWYCDIA. No Necktie Damage. > OK on this --^ I got "Do What You Said You Would Do". But the rest I couldn't get, please help. I couldn't get RAS either. Tim > Murphy's Data Constant: Data will be damaged in direct proportion to its > value LOL! :-) Tim -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -

Sadagopan, Sathish, CFCTR wrote: > > Take a look at this slightly dated paper > http://www.oracle.com/st/o8collateral/html/xoustwp5.html That is VERY dated. As we speak (write?) IDS 7.30 is the base code for the current versions of XPS (8.xx) and UDO (9.xx) so all of the features of the base IDS engine missing from the first releases of XPS and UDO are currently available including replication and stored procedures. IDS 7.31 is only days away from release and it adds some XPS features to the base engine. Informix's stated goal is to have all three "options" using the same code with the various feature sets simply enabled or disabled by the end of 1999. Currently I do not think that there are any features not specific to each option which is not available on all three. The real reason that Oracle 8 provides all of the features in a "single server" is that Oracle has NEVER been a single server. There is a separate executable that extends Oracle locking across multiple nodes in a cluster or MPP box creating the illusion of a parallel clustered server. However the shared locking and shared disks used to implement the protocols reduce performance and tie the nodes together so closely that if one crashes the entire network of servers will crash and have to be restarted. If IDS/XPS loses a node the server stays up and either another node takes over it's disks or a replacement instance is brought up on another node to replace its functions, depending on how you have set things up. Tim Shaeffer who works with XPS all the time can address this better than I. Similarly Oracle 7/8 parallel server technology is based on bringing up additional single threaded processes which can be borrowed by the user's database access process to parallelize queries. Informix defines N CPU VPs, each of which is multi-threaded, which work together to perform queries in parallel. With Oracle more users == more processes. With Informix more users means more threads and not even OS threads to take up system resources, Informix manages it's own threads internally. Oracle 8 adds yet another program to implement universal data. Requests for non-native data types or aggregates and structures of native data types are handled by a separate process. Oracle Data Cartridge packages are each another process with which the user's engine process must communicate and the parallel server modules are not used to aid this. Informix Universal server handles aggregates and structures of native types in it's own code using full parallelization. DataBlades are implemented as shared libraries which can optionally be executed out of the normal CPU VPs or, since some datatypes may involve external data or communication which can slow the engine, you can dedicate one or more special VPS just to handle those datatypes that may cause problems. DataBlades can even be written in such a way that they can be parallelized by the engine, the programmer just needs to follow the rules. Finally there are the outright untruths. Oracle states, for example, that UDO does not use the same administration tools and commands - FALSE, UDO and IDS base will tend to grow apart over time - actually they are growing together, UDO has no indexing of objects and collections - FALSE indeed in addition to whatever built in indexing capabilities IDS provides UDO allows one to create DataBlades with object specific or even generally applicable indexing schemes, UDO does not cache objects - FALSE only BLOB data stored separately from its base row are not cached including objects, shared nothing architecture is less efficient than Oracle Parallel architecture with shared disk and shared locking - the benchmarks belie this one, faulty datablades can crash the entire engine - actually since 7.3 and the latest UDO version only an affected VP will crash the engine as a whole will remain online. Since Oracle Data Cartidges are separate processes Oracle claims that you can run them on the database server, a middle ware server, or even on the client machine where ever it is most efficient. Personally I find that that means that Objects are truly just tacked onto the top of Oracle in the same way that many of us tried to store object data in out relational databases several years ago before Universal server technology was created by Informix.

"Art S. Kagel" wrote: > > Sadagopan, Sathish, CFCTR wrote: > > > > Take a look at this slightly dated paper > > http://www.oracle.com/st/o8collateral/html/xoustwp5.html > > That is VERY dated. As we speak (write?) IDS 7.30 is the base code for > the current versions of XPS (8.xx) and UDO (9.xx) so all of the > features of the base IDS engine missing from the first releases of XPS > and UDO are currently available including replication and stored > procedures. IDS 7.31 is only days away from release and it adds some > XPS features to the base engine. Informix's stated goal is to have all > three "options" using the same code with the various feature sets > simply enabled or disabled by the end of 1999. Currently I do not > think that there are any features not specific to each option which is > not available on all three. > This is news to me. XPS 8.2x has a somewhat different architecture from 7.x, many different SMI tables. Functionally to end-users they don't see any difference, except with a more limited SQL. It does diverge greatly with the coserver_id all over the place. This is one of those items that I think makes merging engines difficult. For example, if you don't buy the added functionality of XPS in a "single engine" of the future, you end up having a lot of dormant things that need to be switchable. The same will be true for the features of IUS being dormant if they aren't used. For example, the HP loader is different, there is no onmonitor for XPS. ( Of course I never used onmonitor anyway, so nothing lost there. :-) Lots of other differences. There are SQL extensions in XPS that do not exist for 7.x and visa versa. We went through training and could not do certain SQL statements available in 7.x. Is this a problem? Not if you keep things simple. Which is what XPS is all about, and that's why it is so elegant in construction. If you build your DW properly with current DW rules, XPS is perfect. It is not for amateurs, and it doesn't do certain things 7.x can do, and for sure it doesn't do 9.x stuff. > The real reason that Oracle 8 provides all of the features in a "single > server" is that Oracle has NEVER been a single server. There is a > separate executable that extends Oracle locking across multiple nodes > in a cluster or MPP box creating the illusion of a parallel clustered > server. However the shared locking and shared disks used to implement > the protocols reduce performance and tie the nodes together so closely > that if one crashes the entire network of servers will crash and have > to be restarted. If IDS/XPS loses a node the server stays up and > either another node takes over it's disks or a replacement instance is > brought up on another node to replace its functions, depending on how > you have set things up. Tim Shaeffer who works with XPS all the time > can address this better than I. > The disks become unavailable on the down server if an XPS node crashes. The data appears to simply be unavailable to the user. This stays sacred to the religion of "shared nothing", the most scalable approach in the known universe for parallel systems. Once the down-server is restarted, the data is available once again, and everyone sighs with relief. 8.3 provides better restart/stop features than earlier versions, to address down-servers, but I've not seen 8.3 just yet. I've also heard that new versions coming out will be moving away from the switch-pipe and back to ethernet. IBM has their ESCON pipe for AIX, but it causes lots of problems, is hard to diagnose, and requires Kerberos, yet another thing to break and cause problems. The key to the future of XPS will be in a simpler high-speed pipe that connects the coservers together. This is an architectural issue, not a technical problem to overcome. The technology is already available. One other BIG problem with XPS is not XPS, but the IECC that comes with it is unusable with 7.x, and is difficult if not impossible to install. Yes, even the ATG guys can't get it to install. It requires way too many network conditions to be perfect. And the IECC for 7.x cannot be used with XPS. Apparently there are several camps of developers at Informix doing multiple versions of IECC. In fact the IECC that comes with XPS for NT will not work with the XPS for AIX! So, needless to say, to this day I use strictly shell scripts and ESQLC to do EVERYTHING! I have quite a toolkit now for XPS ( I don't share these :-). > Similarly Oracle 7/8 parallel server technology is based on bringing up > additional single threaded processes which can be borrowed by the > user's database access process to parallelize queries. Informix > defines N CPU VPs, each of which is multi-threaded, which work together > to perform queries in parallel. With Oracle more users == more > processes. With Informix more users means more threads and not even > OS threads to take up system resources, Informix manages it's own > threads internally. > True as I understand it. Informix is more architecturally similar to UNIX and in a sense is a mini OS inside UNIX. Much of the IFMX "kernel" is modeled after UNIX. Take LRUs, for example, which are a direct offshoot of UNIX LRUs. > Oracle 8 adds yet another program to implement universal data. > Requests for non-native data types or aggregates and structures of > native data types are handled by a separate process. Oracle Data > Cartridge packages are each another process with which the user's > engine process must communicate and the parallel server modules are > not used to aid this. Informix Universal server handles aggregates and > structures of native types in it's own code using full parallelization. > DataBlades are implemented as shared libraries which can optionally be > executed out of the normal CPU VPs or, since some datatypes may involve > external data or communication which can slow the engine, you can > dedicate one or more special VPS just to handle those datatypes that > may cause problems. DataBlades can even be written in such a way that > they can be parallelized by the engine, the programmer just needs to > follow the rules. > Oracle's method of more-is-better is actually fueled by the increase in middleware products, many to this day that I don't understand the need for. So IT/IS people are conditioned to adding more processes and programs and software. Oracle simple rides the train, and this isn't even a sales issue for them. For us "oldtimers" simpler is better. > Finally there are the outright untruths. Oracle states, for example, > that UDO does not use the same administration tools and commands - Well, the UDO for NT doesn't come with an IECC, but instead a bunch of stuff that doesn't make any sense at all unless you're an object orientated application developer. It's no wonder IUS has a hard time being sold, because it doesn't apply to most business the way it sits today. We have data analysts who would love to write their own user- defined function/data-types, but the grief in getting this set up is very costly in the learning curve. I personally don't have time to administer a UDO server--it'd take t