Scale up or out?

Scale up or out?

At STORServer, we provide only the best of the best. One thing I’m regularly asked is when is Tivoli Storage Manager (TSM) better than Commvault and vice versa? While each has its own advantages, the feature I’m most often questioned about is scaling.

Both products scale well but they do it in different ways.

Scale up

Scale up is the practice of using a single server or system of data protection. In the case of STORServer, our appliance powered by TSM does exactly that. TSM uses a scale up model and can handle petabytes and thousands of clients to the same server. To add capacity, users just increase disk shelves or tape systems, and the server can utilize it with an unlimited potential for growth.

This type of architecture has its pros and cons. The scale up environment does seem to provide a seemingly more simple approach as there is usually only one system to manage and pay for. An argument can be made for less ongoing operational expense because the “one system approach” will use less power, space and resources in general.

The scale up approach can also give administrators some serious limitations. With a single system, it will only function up to the capabilities of the components. For example, once you reach the max of the network cards or processing power, there is no more room to scale up. The scale up model has glass ceilings that are hard to break through, and when hit, administrators are forced to scale out. You can only add so much memory and processor power to your systems. Once you hit those limits, you have to scale out to an additional system. The only other option is a “fork lift” upgrade where a newer, more powerful server replaces the system. This involves downtime and can be very expensive.

While it may not seem important, with the growth rate for the average data footprint growing 60 percent or more every year, it could be a deciding factor in the scale up versus scale out argument.

Scale out

Scale out is the practice of distributing loads across multiple systems to provide processing power. To add the ability to back up more data, additional systems must be added. STORServer’s appliances powered by Commvault are a good example of this. As the load increases, more systems can be added to distribute the load (called load balance).

Scale out solutions handle data growth differently. Each system is designed to handle a maximum amount of processing, and when that maximum is reached, the load can then be distributed to additional systems. For example, when users reach the maximum storage of their data protection solution, scale out would have them install a new server, which adds processor power, storage space and bandwidth to the solution. This addition does not take over the load, but rather shares the load. This means that data growth does not lead to longer backup times or recovery times and deduplication and other advanced features can be distributed across the IT infrastructure. This also means there is no maximum capacity.

On the other hand, scaling out means more systems and operational expenses, such as power, cooling, rack space and even ongoing maintenance costs.

Which one is right? There is no single answer. The only thing that can be done is to study both models and make an intelligent decision on how it will affect the solution, the user’s budget and administration time.

How do you make that decision? Planning. Have a good 3- and 5-year plan, and estimate data growth over that time. Once that’s done, users can start making sizing decisions that will help determine whether to scale up or scale out.