PerspectiveApplied Physics

Directing Data Center Traffic

See allHide authors and affiliations

Science  11 Oct 2013:
Vol. 342, Issue 6155, pp. 202-203
DOI: 10.1126/science.1242906

You are currently viewing the summary.

View Full Text

Log in to view the full text

Log in through your institution

Log in through your institution


The widespread adoption of cloud computing has led to the construction of large-scale data centers hosting applications serving millions of users. Underpinning these data centers are tens to hundreds of thousands of servers that communicate internally with each other at high server-to-server bandwidths that are orders of magnitude greater than their connections to end users. Today's data centers consist of racks of 20 to 40 discrete servers, each configured with 8 to 16 CPU cores, hundreds of gigabytes of memory, and potentially tens of terabytes of storage. To meet cost and energy scaling requirements, a new data center design will be required in which a rack of multiple, discrete servers, including the top-of-rack network switch, is integrated into a single chip (see the figure). These integrated “rack-on-chips” will be networked, internally and externally, with both optical circuit switching (to support large flows of data), and electronic packet switching (to support high-priority data flows).