Slashdot Mirror
Open Source Database Underdogs
implex writes "Interactive Week had an article called "Data Underdogs" which they compare offerings of present Open Source Databases with present commercial offerings. In one part they mention ...On the other hand, MySQL developers now have a much-needed transaction management system: NuSphere last month made its Gemini transaction manager for MySQL available as open source code on mySQL.org, a site that the company recently launched. Complicating matters, though, is NuSphere's blood feud with MySQL AB, a Swedish company that runs a competing open source development site for MySQL code at www.mysql.com. No mention of the fact that MySQL AB actually created the product was interesting."
Well, when people talk about market share it is usually by revenue, and given the low price of Open Source databases I'm actually surprised it's even as high as 3%.
But 3% of revenue might actually add up to a fairly substantial proportion of installations.
I don't think you're likely to find that happening on the commercial level. It was so long ago I don't even remember where I read it (though it's the sort of thing that a /. poster might say), but I once heard it said that acknowledging bugs is probably a great way to tick off your investors.
"What, there's a flaw in your product?"
"Yes, but we can fix it pretty easily -- have the bug fix out tomorrow..."
Whereupon the non-technically-inclined investor writes the developer's comments as a sign of weakness and sells. I think that's also why commercial bug fix releases are such a big deal -- probably Microsoft's other reason for cramming all kinds of new features into their service packs.
/Brian
Instead of focusing in on the available solutions, the article simply spits out the idea that MySQL and Postgres are weak pretenders.
The FACT is that these databases are excellent solutions to a large number for MOST database problems. Sure, like all DBMSs, they can always use more features. But I don't want my DBMS to turn into an uncontrolled monster like Oracle.
For 99% of the applications out there, Postgres and mySQL fits the bill. If you're doing large scale distributed payroll using SAP, then I suggest you go with a big name...
But if you're an average-sized business, Postgres is a full-featured solution today. It is an inexpensive, fully-capable solution.
If you're into writing Oracle PL/SQL, a proprietary procedural extention to SQL, go with Oracle. (Note: PL/SQL doesn't work with Sybase or DB2 or anything else.) If you're into TransactSQL, another proprietary SQL extension, go with Sybase. Once you get into TransactSQL, you'll NEVER migrate out without expense. In fact, my shop, an Oracle shop, doesn't PERMIT developers to use the PL/SQL ewxtensions. We learned our lesson after migrating from proprietary MS-SQL-Server extensions to Oracle!
And if you need a big company to support your 20,000 person payroll, go with IBM's DB2. Again, another fine DBMS.
But for the average shop? Save your money and go with PostgreSQL.
What's the problem here? We've got and open source database that's being developed in two somewhat different directions by NuSphere and MySQL AB ... seems like in the end it'll lead to two different, but each (for their intended applications) excellent products. I just don't see a problem.
Maybe I'm just being dense (first reply: "Yes, you're fscking dense!")
You are evil. OSS can accomplish anything ... well, web browsers are hard! ... I'm sure a database is easy though!
One of the things that I like best about PostgreSQL is the fact that the developers are brutally honest about the software. The core PostgreSQL developers have always been quite frank about which parts of PostgreSQL were ready for production, and which parts were kludges, or were largely untested. The problem with commercial databases, even good ones like Oracle, is that the people who know where the rough edges are aren't talking about them. That sort of honesty goes a long way towards building my trust.
PostgreSQL has an amazing featureset, especially considering its price. I think that fairly soon Oracle is going to wake up to the fact that the database is becoming a commodity market, and quite frankly, they aren't likely to be competively priced.
Actually, I've noticed a growing trend in the crowd of programmers I run with is not to care about the implementation platform.. SQL is SQL, and you don't do transactional based stuff on the platform anymore, you abstract away from that into your middleware, and do things the way you want to. In REAL LIFE programming, you often don't have the chance or opportunity to spec out what your back-end platforms are, you have to deal with what's given, or what's legacy at Company XYZ Inc. You also are often dictated a programming langauge for the project whether it be java, c, cobol, perl, python or whatever. The real value lies in being able to adjust to whatever the PROJECT calls for, and being able to implement on just about any platform you need with strong good design patterns. MySQL and Oracle both do exactly the same thing from my point of view, hold data in a relational format for storage and quick retrieval. Putting too much logic into the database only serves to slow things down in the long run. Nasa Switches from Oracle to MySQL shows us why Oracle putting all those bells and whistles in their product may lead to a weakening of their marketshare ultimately. The fact is, bells and whistles cost memory and processor, and there's a balance between the two that Oracle seems to be blithly ignoring.
Did this number strike anyone else as too low? This is the first time I have seen any percentages on OS DBs. Alright, 3%, but of what? Does this mean that less than 3% of all bytes stored in any database are stored in an OS DB? That may be, but I cannot believe that less than 3% of all databases running are OS.
The thing is, when I started studying computer science, we only ever worked on mSQL boxes. My first job used Postgres and the second used mySQL. Anything I work on myself is mySQL. Granted these jobs have all been webrelated, but when you think about how many ISPs offer mySQL/Postgres preinstalled (not to mention linux distributions, if that counts), 3% seems rediculous.
Well I suppose MS Access runs on one or two computers out there... that might raise the non-OS score.
-Kraft
Live and let live
No. The point is that the design of a great DBMS takes a lot more unity than the large-scale projects OSS has tackled previously. In a DBMS, there must be an internal set of standards for everything from datatypes to join optimization logic.
Databases just don't lend themselves to fragmented development the way operating systems do. Frankly, I'm skeptical that an OSS project could (using current development practices) pull together and produce something as capable and stable as DB2 or Oracle.
I've moved a major (~8GB) database from DB2 to PostgreSQL for a client. It runs faster and is easier to feed and administer now. I'm in the process of moving a similar-sized database and app from MS SQL Server to PostgreSQL for the same reasons (plus the openness of PostgreSQL and Linux).
I really like DB2, it's very powerful, robust, and scalable. But it requires a fair amount of admin expertise and time. Not so much as Oracle, but much more than PostgreSQL.
What, frankly, suprised the heck out of me was the fact that nearly all of my queries (this is an audit system, OLAP, not OLTP) ran between two and four times faster under PostgreSQL. That adds up pretty quickly!
As far as the application development support infrastructures, I'm not really sure what is meant by that. The current implementation of stored procedures in PostgreSQL falls short of what DB2 provides, I'll grant. But support for C, Java, Perl, PHP, Python is all there. It's a pretty high-speed/low-drag setup, IMHO.
The set of problems for which PostgreSQL is the best solution is expanding pretty rapidly. I won't pretend that it's the be-all RDBMS, I don't think such a thing exits. I would say that it's worth a serious look for many situations.
Gordon.
He that breaks a thing to find out what it is has left the path of wisdom.
-- J.R.R. Tolkien
We did recently quite a bit of Perl development using MySQL and InnoDB tables, and they worked (surprisingly) well. Having transactions (finally!!) in MySQL is a huge blessing.
Somewhat related...while the article mentions that MySQL and Postgres don't have the large application development support infrastructures that the bigger commercial database have, they can be a lot quicker to prototype and develop with because of their relative simplicity.
We're in the middle of migrating our application to DB2 on RS/6000, and I have to say I'm missing MySQL's simplicity of administration and configuration...you can try out a lot of new ideas quickly with MySQL, whereas a big chunk of our time at the moment is spent poring over DB2 manuals for obscure command switches and SQL options (the LOAD utility can be a barrel of laughs for newcomers)...of course if our DBA was a little more competent, but that's a different story :-(
(And yes I do realise DB2 is much more powerful/robust...I'm talking about ease of development and rapid prototyping!)
It's a pretty tall-order, but it lives up to it. Our internal benchmarking tests for our application purposes show it to be about 7x faster than an identical PostgreSQL 7.1.2 solution. I've seen reports on the mailing lists that it can be up to 18x faster. You also get the simplicity and maturity of MySQL. The InnoDB benchmark page has their own benchmarks, which pretty much mirror what we've seen internally.
Just a quick look at the benchmarks link tells me that they have fsync turned on on Postgres. What exactly is fsync? Every time Postgres touches the disk, it sync()s. Slow? Hell yeah. But you won't lose data in the cache. It's turned on by default.
I realize that Postgres isn't the fastest in the world, but it's not 7x slower on 100k inserts. That's just bad benchmarking. Deceitful even.
If fsync is not on, I apologize. However the link mentions no performance tuning other than buffer pools and log buffers. If Postgres is defeated by 7x (18x?!) in a fair test, I'll concede. However this looks like the MySQL testing benchmarks on mysql.org; bullshit, plain and simple.
One is tied up in this contractual mess, but the other, InnoDB has no such issues, and may even be faster for many purposes.
That might be the understatement of the year. InnoDB touts itself as the "fastest disk-based database" currently on the market. It's a pretty tall-order, but it lives up to it. Our internal benchmarking tests for our application purposes show it to be about 7x faster than an identical PostgreSQL 7.1.2 solution. I've seen reports on the mailing lists that it can be up to 18x faster. You also get the simplicity and maturity of MySQL. The InnoDB benchmark page has their own benchmarks, which pretty much mirror what we've seen internally.
Of course, MySQL has other drawbacks, namely that it doesn't support triggers or table inheritance or some of the more complex nuances of standard SQL, but the 95% of stuff it does have is very fast, and the other 5% can be handled in code. MySQL isn't popular because it's open-source, though. It's popular because it's good, free, and most importantly, extremely easy and intuitive to use.
One problem with proprietary DBs is that their docs will steer you toward non-standard SQL even when standard SQL will work. For example, Oracle will teach you to use NVL and Sybase will teach ISNULL, when COALESCE works in both databases.
The solution is to develop with PostgreSQL regardless of what your deployment DB will be. Their docs favor standard SQL. The code you develop will work with the proprietary DBs as well.
Your DBA was running around tweaking all day because he was supporting a poorly designed application that was completely unoptimized.
Because so many developers have a notion (prominently displayed in this story) that "SQL is SQL" many apps running on a database run horribly inefficient queries that bog the database down.
Studies have consistently found that80-90% of database tuning needs to take place in the application. The database tuning portion mainly consists of tweaking memory and parallel query options.
Conformity is the jailer of freedom and enemy of growth. -JFK
Companies keep everything on database systems. Hundreds of geek-hours must go into the design of a database application for a company. Whatever system a company chooses, they must be reasonably sure the system will:
- almost never fail
- be supported by a stable company and
- integrate well with other systems, into the future.
A smaller price tag may be a good start to target smaller companies that don't rely heavily on database applications, but the reason Oracle can charge $15k/CPU for 9i Standard: the reputation is worth it.Yeah, all the companies would immediately say "Let's base the future of our business on new and untested technology! Better yet, let's buy it from an unknown vendor!"
Please. The reason relational databases are still very much the dominant technology is that they work, they work well, and they've been working well for decades.
I only have experience (but lots of it) with Oracle and - unfortunately - SQL Server. Until SQL Server 6.5, I think it sucked compared to Oracle. But now they're more on level, although Oracle has always, and still does, handle huge data warehouses much better.
That aside, I worked for years with a 4TB data warehouse for a major credit card company. It was Oracle (7?) on a Sun E10000 (22 processors, 1GB ram) and it was screaming. We barely used any "advanced" features that were unique to Oracle. But what impressed me was Oracle's support. They had an office a few miles away and would send DBAs over to help out. Our DBAs were excellent, but when it came to very low-level tweaking, these Oracle DBAs knew their stuff. They would mess around with the OS to keep it as efficient as possible. And if there was ever any kind of failure or error, they came over to check it out.
Now granted, my company paid big bucks for the support, but at the moment that sort of support can't be found for an open source dbms. These were highly skilled experts in the database they supported. I realize (partly from the article) that the current goal of open source databases is to grow in the low-end market - smaller systems and such - and I'd bet they'd stand up to large warehouses. But one big advantage Oracle and DB2, and to a much lesser extend SQL Server, have is their support. You can have a highly skilled technician in your office very quickly if you need it, beyond the support of a consultant could provide. I'd like to see that kind of support in open-source companies. That's when I think they'll give closed-source databases a true run for their money... literally.
Developers: We can use your help.
"Databases are dramatically more complicated than any Web server or operating system technology."
The above is a quote from senior marketing director Bob Shimp, from the article. I will give him the Web Server - which is not to say that it is not complex, but likely not as complex as a robust relational database. I cannot do the same for the OpSys. There is a dramatic difference in the levels of complexity between a monolithic single-user non-multitasking operating system (such as DOS) and a multiprocessing distributed parallel asymmetric (etc etc) OpSys. The quote is not grounded in any sort of evidence, and I have serious doubts as to whether the 'marketing director' would have ever encountered a kernel that did not come from a bag marked 'Orville Redenbocker'. It is simply misguided and misinformed, and the general intent seems to be in undermining confidence in Open Source DBs. (... furthering the myth that open source is 'unreliable'.) Threatened? He likely should be.
I'd like to see an object data model (ODM) open source database come into the scene. Now that would cause a ruckus, challenging both the bottom line and validity of the relational model!
MySQL has two table types that support row-level locking and transactions. One is tied up in this contractual mess, but the other, InnoDB has no such issues, and may even be faster for many purposes.
Actually, Interbase can do "hot" backups and has support for replication.
People really ought to look more closely at Interbase. It just works.
Bah ! I remember in the early 80's when big iron buddies used to point ant laugh at dBase II. What they didn't understand, and what some of the big database boys and users don't understand now, is that larger isn't always better.
Databases like MySQL make it very easy for webhosting companies to offer free databases without loosing their shirts or minds. They make it very easy for students to learn SQL. They're also much kinder on resource.
Yes, I'd love to be able to roll-back pooched transactions, but then I have to commit everything as well. Certainly cascades would be slick, but poorly written, they can shoot your foot clean off. Likewise, I can see all the lame support calls coming in because users don't understand the triggers are attempting to maintain referential integrity on foreign keys.
Within a given context, sometimes smaller is suits the purpose better.
healyourchurchwebsite.com - WWJB?
If a database goes wrong (and in Oracles case, my experience is that that's often), and we can't solve it ourselves, we need to be able to get on the phone and speak to somebody who can help. Now, I know that there are companies that offer support for OS DBMS's, but Oracle, Sybase and IBM's round the clock support offering is what i'm after. and getting skilled technicians (possibly the development team itself) involved quickly. OK, so open source offers this as by merit of "use the source luke", but in a corporate environment, this is neither likely or necessarily sensible.
Another, and perhaps more important, aspect to bear in mind (and this is not covered by the article for obvious reasons) is that Oracle, Sybase and DB2 are not the be-all and end-all of RDBMS offerings. There are better, and often significantly cheaper, closed-source offerings out there. One of my current favourites (which I'm working with at the moment) is Clustra - a DBMS that offers 99.999% availability, scheduled and unscheduled, pretty much out of the box, with Linux as their first released OS for the latest 4.1 offering
So, in a nutshell - Open source support offerings need to be improved, but don't rule out the smaller fish in this crowded, and very competitive pond.
Who said databases were boring?
ooooooh! What does this button do? - DeeDee, Dexters Lab.
I don't think you can "abstract away transaction related stuff" from the RDBMS. Please correct me if I'm wrong; but I believe that you simply can not start with a RDBMS that does not support transactions; and slap a transaction processing middleware on top of it to instantly have a database application with transactions.
The standard transactional interfaces that are increasingly becoming popular nowadays (like JTA) depend on the database supporting transactions; most likely in the form of an XA-conformant programming interface.
Zigbee Central: A Zigbee weblog
Nobody seems to have noticed SAP DB becoming Open Source last year, but it too is powerful and reliable and includes full transaction support.
Got time? Spend some of it coding or testing