Database Reference
In-Depth Information
Batch applications
For data concentrated applications, for example, applications that are targeted by the Ha-
doop platform, it is easy to scale the bandwidth and you can simply add more servers to the
cluster to scale out the throughput, given that it is reasonable to get an elevated bandwidth
both through in-memory technology and through disk-based technology by using horizont-
al scaling. The very famous project called RAMCloud has made some arguments such as:
• In-memory technology is cheaper in certain use cases
• Hard drive prices also come down year by year
If you wish to admit each data item more commonly, you simply cannot fill up the disk;
you will throttle the disk IO interface. You can only utilize a small portion of an HDD if
you need high IO throughput, which will definitely take your effective cost per bit up. At
various points it will be costlier than an in-memory solution. If you are going to deploy
your own infrastructure where you are bearing the entire infrastructure cost, it may make
sense to use the in-memory technique for batch applications. However, for hosted cloud en-
vironments where you are willing to pay for the actual storage you use, an in-memory tech-
nique such as AWS DynamoDB may be the right candidate for your batch applications, but
again it's based on your application and cost.
