CS team develops RAMCloud

Nov. 3, 2011, 3:00 a.m.

A team led by John Ousterhout, professor of computer science, recently completed a vital stage in the development of RAMCloud, a key value data storage system that will improve performance levels of current storage technology tenfold and open up possibilities for future networking sites and web applications.

RAMCloud builds upon existing large data storage systems, like those used by Google and Facebook, and aims to enable users to store and recover more data at lower latency speeds. According to Ryan Stutsman, a computer science graduate student working on the project, this type of technology is a relatively new experimentation.

“Using RAM for this kind of approach is only recently becoming practical as the size of RAM has been scaling up over the years and it’s now able to hold large data sets,” Stutsman said.

Ousterhout came up with the initial idea for the project in early 2009, after looking at how networking and search engine sites were storing memory and data. With the help of Mendel Rosenblum, associate professor of computer science and electrical engineering, the idea grew into the current project, which is based on both professors’ interest in exploring a low latent storage system.

“We made the decision that, if we were to do a company, we might have to do something different. But as research we would try to build something that would be fast and have the lowest latency possible and start from there,” Rosenblum said. “That led to the idea that we would put everything in memory and call it RAMCloud.”

The project’s planning stage lasted for nine months, until late fall of 2009 when up to 10 students and numerous faculty began directly contributing to coding and organization. Stutsman indicated that while the program is far from being finished, a basic form exists.

“The RAMCloud software has basically been able to service basic operations and has been usable in a testing way for about a year and a half now,” Stutsman said. “As far as storing real data in it and supporting a legitimate web application, we just put in place one of the most important pieces to head in that direction . . .we’ve developed a crash recovery system.”

The crash recovery system allows notes that have crashed to quickly “reconstitute the memory that was part of that crash host,” according to Stutsman. He indicated that in establishing this system, the team has overcome one of its most significant obstacles.

According to Stutsman, the major challenge in using RAM for the storage system is the volatile nature of RAM itself. Whenever a machine crashes or power is lost, software based on RAM loses all the data stored in it. With the recently completed crash recovery system, however, the project resolved this challenge.

The RAMCloud’s magnetic disk foundation surpasses the cost issue associated with RAM technology. While traditional data center storage systems solely use magnetic disks, the RAMCloud software can offer the durability and cheap price of disks with greater performance levels and lower latencies of RAM. This combination makes the RAMCloud even more accessible to users for large networks.

David Erickson, a graduate student in computer science involved in the initial exploration of the project, said that the RamCloud’s success would play an impacting role in modern technology.

“If RAMCloud excels and can meet its goal, then dispatch steps could potentially be much quicker for just about everything in the Internet,” Erikson said.

The overall goal for the project is to allow for the creation of new applications, without the creation process being hindered by slow network speeds or limited space. Stutsman said he believes technology is entering a new era in which large data center storage systems will require the speed and efficiency that RAMCloud aims to provide.

Stutsman said the next stage in the process will involve polishing additional features that would make this goal all the more attainable.

“We think this project can slingshot applications into the next level of data intensity,” Stutsman said. “We think that there will a lot of interesting applications that can get at their data and access data and basically provide more features and a more interesting experience to users in general.”

Ileana Najarro is the Managing Editor of News at The Stanford Daily. She previously worked as a News Desk Editor and Staff Writer.

Login or create an account