24x7 enterprise databases

Hi, I am Noel Yuhanna, an analyst at Forrester Research. Today I will be talking about 24x7 enterprise databases. What does it mean? Well, databases today are very critical in any organization, in banks, in insurance companies. And you have to ensure availability of these databases. How do you ensure it? Databases could fail.

There could be multiple reasons why a database fails actually. Three of the most common ones we know of are hardware failures, software failures and human error.

Software failures account for 30% of all failures whereas human errors account for only 15% of all failures. The biggest portion of these failures are accounted to hardware failures, which is 55%.

Let's look at the reasons why hardware failures can happen. When an application goes down, there can be reasons why it goes down. It could be a network failure or it could be a server failure or a storage failure.

There could be four different solutions available, to help you with availability. One is replication. In a typical environment you would have an application connected to a database and to a server and you may have another database and another server available in different locations. And what you want to do is replicate data. Whenever transactions are coming into to one server you want to replicate to another server, to make it available so that if one server goes down, the second server can make it all available.

Well there are some limitations however with this technology. All data might not be available to the second server. There could be some latency there. So you may lose transactions around these applications. So it is a good technology, easy to implement, but may not serve the purpose of complete availability.

The second solution we know of is shared nothing clustering. Now this is a good technology, which you can have a segregation of data, which really splits the data or breaks down the data into multiple chunks. Some data may be stored in this database, some data may be stored in this database as well and both of them together deliver information toward the application.

Now there could be the possibility that with one server and application or the database may go down. What do you do? Well your data may not be completely available, so there are some limitations as well with this technology.

The third technology which we know of is shared disk clustering. How shared disk clustering actually works is that one application connects to a database and to a server and to a storage. And the second server is just a standby, just being available if at all necessary. And there is communication going on between the server and server to understand where does the issue lie, if the server is available or not. If the server goes down for some reason this other server will pick up the data and make it available to the application. So this is a good technology, but may again take you some time to get to that situation. It may take five minutes to seven minutes to recover the data and make it available to the application.

The fourth solution that we know of, which is actually an evolution of the shared disk clustering is called the multi-node-clustering. How the multi-node-shared-disk clustering works, is that the application interacts with the database which has multiple servers associated with it and all these servers tie to a common storage. So all of them deliver the same type of information available to the application.

Now the good thing with this technology is that if at all a server goes down, your other servers are available, and serving the application. In fact you can have more servers if you require, to make it available as well. You can have 10, 16, 24 nodes, we call it nodes available in this environment. So this is a good technology which serves a better purpose. It also delivers to you availability for the application as necessary. And we see a growing trend where we see multiples of these low cost servers acting as a shared disk clustering solution.

To summarize, we looked at why databases fail. One of the reasons is hardware failures. We also looked at four different solutions including replication, shared nothing clustering, shared disk clustering, and multi-node shared disk clustering. To ensure availability of 24x7, the best solution that we know of is multi-disk node clustering. We believe multi-node shared disk clustering delivers the highest availability.