Today’s data centers have ever-increasing workloads placed upon them, resulting in the wildly expensive cost of operations and maintenance in addition to supplying the thousands of watts of power they can require. This is the question highlighted in the Oct. 11 edition of the journal Science by Electrical and Computer Engineering Department Chair Shaya Fainman and Center for Networked Systems Associate Director George Porter, both faculty at UCSD. In order to accomplish energy and monetary conservancy in future centers, Fainman and Porter discuss one option which would change the current data center design into racks-on-chip. Rather than large racks of servers arranged in conjunction, individual chips will have “racks” that act like a miniaturized server.
“The idea behind racks-on-chip [is] to take the processing power and memory of individual servers in these data centers, and start to integrate them into highly dense packages, reducing the overall power and cooling,” Porter said in an email interview. “By increasing the density of these internet data centers, we have the potential of doing more computing in less area, using less power and cooling.”
Density on chips refers to the amount of information that can be stored on a given surface area. Increasing the density, and therefore the quantity of data able to be contained on the chips, is necessary in order to make the racks-on-chip design a reality.
Many online applications in particular work off data centers “running in the cloud,” according to Porter. These proposed future data centers would not only need to have increased density, but would also have to network the racks-on-chip together. Thus the final product would help solve the current cost and power issues while adding performance, since more computing may be done with this new design.
The data center on campus is considered inadequate to current standards and in the process of modernization.
“Our current data center is a small, retrofitted printing facility,” said Mark Redican, director of Communication Resources for Information and Educational Technology, who is leading the project to revamp the data center facility. “It was not architected to provide the efficient power, cooling and failure resistance expected in a modern data center,” Redican said.
The improvements, once complete, will fix energy efficiency and help to guard the most important applications and systems. The increasing demand placed upon the center will then be supported by a data center built to tackle the needs of UC Davis.
“As the amount of data increases, we will need to either build more data centers to handle it, or pack more compute power into existing data center spaces. Racks on a chip would be one solution to providing denser computing power per square foot within existing data centers,” Redican said, when asked about the proposed racks-on-chip data center design.
Racks-on-chip is one potential solution to cut down the operating cost of data centers, and could be a viable option once the hardware challenges are conquered. According to Porter, the density and networking requirements for the racks-on-chip and the time and effort to finally tie everything together to build a working model would take an estimated “decade.”
“Data centers will always be limited by cost and power usage,” said Bevan Baas, an associate professor in the Electrical and Computer Engineering Department at UC Davis. “There is no problem finding applications that benefit from additional computing and storage resources and they will saturate the capabilities of any hardware — just as what happens when I buy a new desktop computer and within a short time the new applications make it as slow as my last computer.”
Looking toward the roadblocks ahead for racks-on-chip databases, there are many questions regarding the full design and networking of the chips. The journal publication explains the puzzles needed to be solved; the design is not a guarantee of a working model, but rather a suggestion for what researchers can begin to investigate.
“It is not clear whether it will be possible to integrate massive data storage on [a] chip and I think that would be the major obstacle to implementing a rack on a chip,” Baas said.
But are 10 years of dedication to this design worth it?
“Yes! We’re just starting to see … the potential of large-scale cloud computing, and as we’re able to do more computing in those data centers, with more data and information, we’ll begin to see more interesting applications and services,” Porter said.
Take everything we do every day via the internet and on our many computers and devices, be it Googling, shopping, or conducting research. We rely on the amount of computing available to complete all of these tasks. By working towards the goal of performing more work, more computation within existing datacenters and in our technology, we will be able to complete tasks unheard of today.
“If we’re to tackle big challenges like climate modeling and genetic research, we need to harness much more computing power than is practical today,” Porter said.
