Cloud computing has the advantage of being much more flexible than similar hardware-based services. However, cloud services tend to fall behind when it comes to database-intensive applications due to limitations in hard drive speeds. Updating data in a hard drive is the limiting factor for most computers nowadays, as the process is limited by the speed of the stick that is writing the information to the disk.
MIT’s news article, “Making Cloud Computing More Efficient,” written by Barzan Mozafari, explains that “updating data stored on a hard drive is time-consuming, so most database servers will try to postpone that operation as long as they can, instead storing data modifications in the much faster – but volatile – main memory.”
At the SIGMOD conference, MIT researchers will reveal algorithms used by a new system called DBSeer that uses a “gray box model” that should help solve this problem. DBSeer will use machine-learning techniques that will be able to predict the resource usage and needs of individual database-driven application servers. Cloud computing servers are often divided up into multiple “virtual machines,” which are partitions of servers which are each allocated a set amount of processing power, memory, etc. DBSeer will hopefully be able to predict a database’s unique needs and idiosyncrasies so it can predict whether or not it is viable to allocate additional resources from other partitions to solve a task. If a virtual machine is just sitting there idle, DBSeer will assess whether or not it is prudent for that virtual machine to continue sitting there, or spend its allocated resources to complete a task on another partition.
Ultimately, this will allow servers to be much more efficient without further investment in hardware. This trend that follows with Big Data is really getting computer scientists to question if there are more efficient ways to handle our problems with the hardware that we have. It is all about maximizing productivity by questioning our own methods, rather than simply investing in more hardware.