« WebSphere 6.0.2 ships | Main | Difference between startup beans and XD partitioning »

July 29, 2005

Oracle, spoons and toasters...

Oracles new grid ads are great. A 390 and a bunch of pizza box PCs. They have got to be kidding. It's like comparing a spoon with a toaster. Yep, nothing in common but hey, they both are silver! 390s run much of the worlds OLTP right now and for the customers that own them, the majority swear by them. They run mixed workloads with business based goals and happen to run databases also. Eventually, grids of less powerful computing may be able to replace them but it'll be a while yet. WebSphere XD offers something similar for J2EE workloads but still is not as sophisticated or as end to end as what Z/OS WLM offers.

I think the bigger story with the ad is that Oracle really wants to sell RAC to a market which doesn't need it. I said it, there is a very limited market for a RAC like server. I think most databases will happily run on servers with 4 processors or less and it's about to get worse for database vendors in general including our own DB2.

My theory here is that as caching middleware moves more and more database like function up in to the application tier, the need for horizontally scalable databases is going to vanish. Products like our own WebSphere XD ObjectGrid or its competitors must scare the pants out of the database vendors when they look at the amount of processors needed to host a database that is being used by applications leveraging this type of middleware.

The majority of customers simply do not need huge database boxes to host the data. More and more of the MIPs currently burned in the database tier will move up in to cheaper tiers running the new generation of caching middleware and applications will never have run faster.

With dual core processors arriving, the power in even modest dual CPU (quad core) blade servers is pretty incredible and in conjunction with this type of caching middleware is capable of awesome performance.

Availability is the other argument for RAC but as many in the business know, RAC isn't what its cracked up to be but we all agree Oracle has great marketing. Solutions like IBM DB2 HADR which implement a peer to peer primary/slave database cluster provide as good as or better availability with a much simpler and easier to test/implement configuration. I've even seen customers just using basic replication to make a hot standby database are good enough for many customers, even ultra high end ones, you'd be surprised and it's fine depending on the business requirements to do it this way.

The dawn of this new caching middleware has arrived and the vast majority of customers don't need a horizontally scalable database for either performance or availability.

Customers need to educate themselves on simpler solutions for database scalability and availability. Lets just partition the data, half the data on database A, half on database B. Lose a database then you can still access half your database on the second database which is independant of B. A fails, B still works, guaranteed! This arrangement may be acceptable for your business. When A restarts, A's data is available again. If the databases were DB2 HADR this takes a very short period of time and is a lot less hassle than configuring SANS and very networks and RAC and it's a lot cheaper. The main point here is that when A fails there is absolutely no interruption to B, NONE. The application just has to use the right database connection for the transaction, something XD has explicit support for.

This simpler approach will yield more reliable systems simply because they are easier to setup and test. I've had a number of customers bang this in to my head with a sledge hammer :) WebSphere 6.0 ND goes a long way towards this kind of simplification with its unique exploitation of commodity NAS technology as a way to provide high availability. This approach coupled with increasingly sophisticated caching middleware like XD ObjectGrid or its competitors should make customers looking at horizontally scalable databases think twice before jumping in. There are cheaper solutions that will deliver very high performance at a much lower cost and are simpler to deploy and work on your existing infrastructure.

July 29, 2005 | Permalink


I wonder who they are targetting.

Mom and Pop will not buy a mainframe or a grid server or oracle for that matter.

Small and Medium business same thing.

Big business they know what they are doing and will think this as a joke.

So they are going after mindshare? still difficult to comprehend..

Posted by: | Aug 1, 2005 9:34:19 AM

Post a comment