You're looking at the price/performance ranking - for oltp applications. SQL Server does have the best numbers there. However, this doesn't reflect best performance for oltp (where they hardly show up at all), and certainly not for searches/reporting (which are reflected in tpc-h).
And the only parallel features that I'm aware that they support are in using a set of distributed views to provide a single image of a set of independent servers. This is just a trivial implementation of some of the parallel features that db2 and informix use. It really isn't even in the ballpark for this kind of functionality.
OSS databases can work in enterprise deployments - you've just got to have to be prepared for lower availability & reliability numbers. Often that's not a big deal, and often it can be addressed through redundancy (which is easier with $0 cost for postgres, and generally low cost for mysql).
I've been a dba for sql server, db2, oracle, and postgresql. And actually db2 8.1 is pretty easy to work with. Still clunky in some ways, and not as easy as sql server, but much simpler to admin than oracle (which has 450 page install docs and 800 page backup docs).
> If you feel that hosting the CLR in process with SQL Server is a bad idea, how do you feel about hosting a JVM ?
I prefer to keep component functionality lean. So, I've got no problem with JVMs and CLR - but I just absolutely don't want it in a database server.
Each of these technologies is sufficiently complex on its own, and far more complex when integrated by a vendor.
> SQL Server 2005? is going to be great. However, if I had to choose the *best* database I would go with
> Oracle without a doubt. Every tool other database manufacturers are trying to mirror
> generally come from Oracle.
Ah, no. Most of the BI functionality being implemented in RDBMS' over the last 4-8 years was first implemented in teradata & then informix.
And "Best" makes no sense without context. Oracle is very powerful, but relies heavily on experienced and available dbas, and is the most expensive product out there. Best for some folks might be a less expensive product or simpler product to manage. And as far as performance is concerned both teradata & db2 beat it out at the extreme top end.
> As long as you can accept the limitations of MySQL, it's perfectly usable.
Don't forget the design philosophy - in which everything inconvenient is given a default operation.
Thanks anyway - my only interest in mysql is if I need to walk someone through a windows installation of a free database. Other than that, I'd go elsewhere to avoid loosing money on data corruption.
> MySQL is faster and lighter weight than PostgreSQL in my experience.
It is lighter-weight, but in mixed-workload tests I've done (using innodb) postgresql was faster.
well, even though 2000 is a decent product - it still has quite a few obvious enhancements needed (exception handling in stored procs for example).
The direction they're headed with yukon however, is to push more.net stuff into the database. Yuk. I've always avoided their proprietary ado/vb/etc stuff in sql 2000, and.net is just more of the like.
Anyone who takes advantage of any of that stuff is stuck with sql server forever.
You're right - materialized views are pretty useful in OLTP as well. Though I only use them when desperate - since i prefer to do more transformations on data than can be easily done with the MV.
> There are a whole host of other features that Oracle has that nobody else even comes close:
> function based indexes, index organized tables and advanced queueing, wait based monitoring, very > good execution plan selection plus fine grained optimizer hint overrides, etc., etc., etc...
DB2 seems to be coming along pretty nicely with its 8.1 version. I'm building a warehouse with it right now - and its multi-dimensional clustering combined with partitioning across multiple machines results in a very fast system. The only thing I miss from oracle is transportable tablespaces, but with my reload strategy it isn't a killer.
Haven't tried teradata - I assume that product line is pretty good as well.
But there's a huge gulf between the open source solutions, sql-server and the databases that can manage terrabytes of data.
How often do you see case studies in which management says that something they tried really stank? The only time you'll see that is when they intend to litigate - which is rare.
Most of the time these "real world results" are produced by ambitious executives that want more publicity within their own company: they never want to take responsibility for a failure.
Fortunately for the open source movement, it's gaining popularity - so you'll tend to see more of these in support of it. It'll be almost impossible to find one that opposes it (even though it is sometimes a failure).
> You don't have stored procedures, triggers, or views, but as an Oracle DBA who has 12,000 lines > of proprietary Oracle-PL/SQL on just one of my three databases, I can say that they cause more > problems than they are worth. A nightmare to write, debug, and maintain.
Sure stored procedures can get complicated, then again so can anything if you're fool. Most that I see written today are extremely simple. But anyone who's having nightmares from the complexity of views needs to step away from the keyboard slowly.
> But never the less, views, stored procedures, and all those other features are due out in 4.1 and 5.0.
Yep, and the core functionality *almost* works reliably. Really, views, etc are essential in a database - but they need to first clean up the silent errors in their core functionality.
Don't take my word for us - read their manual (especially the "design issues" section), then apply all that invaluable Oracle dba experience to a few tests. You might be amazed at how often mysql corrupts your data. You just never realized it before because every time it hits something difficult, complex, confusing, etc it simply takes some weird default action. Isn't quite so bad for read-only applications, but anyone who writes valuable data to mysql without really understanding the product is a fool.
Oh yeah, and the best part - it isn't free either.
you just need to know where to be careful, since when they fail they won't reliably tell you the way most other databases will.
What they'll do, and what most of mysql will do is to do whatever will allow the operation to complete. So, as long as you know what & when to test - you get can good transaction support out of mysql.
Not a big deal if you know what to look for and don't mind writing that code. Personally, I wish they'd get their shit together and fix this crap since I'm getting very tired of writing this kind of code.
I've seen the licensing fees - and if I remember right, they charge by concurrent user.
So, if you want to support 1000 concurrent users @ $500/user on a mysql-based web portal, wouldn't that be something like $500,000?
Sounds commercial to me anyhow.
Slashdot Mirror
You're looking at the price/performance ranking - for oltp applications. SQL Server does have the best numbers there. However, this doesn't reflect best performance for oltp (where they hardly show up at all), and certainly not for searches/reporting (which are reflected in tpc-h). And the only parallel features that I'm aware that they support are in using a set of distributed views to provide a single image of a set of independent servers. This is just a trivial implementation of some of the parallel features that db2 and informix use. It really isn't even in the ballpark for this kind of functionality.
OSS databases can work in enterprise deployments - you've just got to have to be prepared for lower availability & reliability numbers. Often that's not a big deal, and often it can be addressed through redundancy (which is easier with $0 cost for postgres, and generally low cost for mysql). I've been a dba for sql server, db2, oracle, and postgresql. And actually db2 8.1 is pretty easy to work with. Still clunky in some ways, and not as easy as sql server, but much simpler to admin than oracle (which has 450 page install docs and 800 page backup docs).
> If you feel that hosting the CLR in process with SQL Server is a bad idea, how do you feel about hosting a JVM ? I prefer to keep component functionality lean. So, I've got no problem with JVMs and CLR - but I just absolutely don't want it in a database server. Each of these technologies is sufficiently complex on its own, and far more complex when integrated by a vendor.
> SQL Server 2005? is going to be great. However, if I had to choose the *best* database I would go with > Oracle without a doubt. Every tool other database manufacturers are trying to mirror > generally come from Oracle. Ah, no. Most of the BI functionality being implemented in RDBMS' over the last 4-8 years was first implemented in teradata & then informix. And "Best" makes no sense without context. Oracle is very powerful, but relies heavily on experienced and available dbas, and is the most expensive product out there. Best for some folks might be a less expensive product or simpler product to manage. And as far as performance is concerned both teradata & db2 beat it out at the extreme top end.
Oracle and DB2 both had transactions by '83, so mysql's about 21 years behind by that measure.
> As long as you can accept the limitations of MySQL, it's perfectly usable. Don't forget the design philosophy - in which everything inconvenient is given a default operation. Thanks anyway - my only interest in mysql is if I need to walk someone through a windows installation of a free database. Other than that, I'd go elsewhere to avoid loosing money on data corruption. > MySQL is faster and lighter weight than PostgreSQL in my experience. It is lighter-weight, but in mixed-workload tests I've done (using innodb) postgresql was faster.
yeah, well keep in mind that Oracle & DB2 only got views around...1981?
well, even though 2000 is a decent product - it still has quite a few obvious enhancements needed (exception handling in stored procs for example). The direction they're headed with yukon however, is to push more .net stuff into the database. Yuk. I've always avoided their proprietary ado/vb/etc stuff in sql 2000, and .net is just more of the like.
Anyone who takes advantage of any of that stuff is stuck with sql server forever.
You're right - materialized views are pretty useful in OLTP as well. Though I only use them when desperate - since i prefer to do more transformations on data than can be easily done with the MV. > There are a whole host of other features that Oracle has that nobody else even comes close: > function based indexes, index organized tables and advanced queueing, wait based monitoring, very > good execution plan selection plus fine grained optimizer hint overrides, etc., etc., etc... DB2 seems to be coming along pretty nicely with its 8.1 version. I'm building a warehouse with it right now - and its multi-dimensional clustering combined with partitioning across multiple machines results in a very fast system. The only thing I miss from oracle is transportable tablespaces, but with my reload strategy it isn't a killer. Haven't tried teradata - I assume that product line is pretty good as well. But there's a huge gulf between the open source solutions, sql-server and the databases that can manage terrabytes of data.
How often do you see case studies in which management says that something they tried really stank? The only time you'll see that is when they intend to litigate - which is rare. Most of the time these "real world results" are produced by ambitious executives that want more publicity within their own company: they never want to take responsibility for a failure. Fortunately for the open source movement, it's gaining popularity - so you'll tend to see more of these in support of it. It'll be almost impossible to find one that opposes it (even though it is sometimes a failure).
> You don't have stored procedures, triggers, or views, but as an Oracle DBA who has 12,000 lines
> of proprietary Oracle-PL/SQL on just one of my three databases, I can say that they cause more
> problems than they are worth. A nightmare to write, debug, and maintain.
Sure stored procedures can get complicated, then again so can anything if you're fool. Most that I see written today are extremely simple. But anyone who's having nightmares from the complexity of views needs to step away from the keyboard slowly.
> But never the less, views, stored procedures, and all those other features are due out in 4.1 and 5.0.
Yep, and the core functionality *almost* works reliably. Really, views, etc are essential in a database - but they need to first clean up the silent errors in their core functionality.
Don't take my word for us - read their manual (especially the "design issues" section), then apply all that invaluable Oracle dba experience to a few tests. You might be amazed at how often mysql corrupts your data. You just never realized it before because every time it hits something difficult, complex, confusing, etc it simply takes some weird default action. Isn't quite so bad for read-only applications, but anyone who writes valuable data to mysql without really understanding the product is a fool.
Oh yeah, and the best part - it isn't free either.
you just need to know where to be careful, since when they fail they won't reliably tell you the way most other databases will. What they'll do, and what most of mysql will do is to do whatever will allow the operation to complete. So, as long as you know what & when to test - you get can good transaction support out of mysql. Not a big deal if you know what to look for and don't mind writing that code. Personally, I wish they'd get their shit together and fix this crap since I'm getting very tired of writing this kind of code.
I've seen the licensing fees - and if I remember right, they charge by concurrent user. So, if you want to support 1000 concurrent users @ $500/user on a mysql-based web portal, wouldn't that be something like $500,000? Sounds commercial to me anyhow.