Hunting mice with microsecond circuit switches

Abstract

Recently, there have been proposals for constructing hybrid data center networks combining electronic packet switching with either wireless or optical circuit switching, which are ideally suited for supporting bulk traffic. Previous work has relied on a technique called hotspot scheduling, in which the traffic matrix is measured, hotspots identified, and circuits established to automatically offload traffic from the packetswitched network. While this hybrid approach does reduce CAPEX and OPEX, it still relies on having a well-provisioned packet-switched network to carry the remaining traffic. In this paper, we describe a generalization of hotspot scheduling, called traffic matrix scheduling, where most or even all bulk traffic is routed over circuits. In other words, we don't just hunt elephants, we also hunt mice. Traffic matrix scheduling rapidly time-shares circuits across many destinations at microsecond time scales. The traffic matrix scheduling algorithm can route arbitrary traffic patterns and runs in polynomial time. We briefly describe a working implementation of traffic matrix scheduling using a custom-built data center optical circuit switch with a 2.8 microsecond switching time.

Integrating microsecond circuit switching into the data center

“…In [7], we proposed an approach for scheduling circuits, called Traffic Matrix Scheduling (TMS). In this section, we expand on that initial idea, and then implement and evaluate it in the context of a testbed deployment later in the paper.…”

“…Any remaining traffic is instead sent to the packet-switched network. Using the same randomly generated TDM from [7], we calculate the tradeoffs in different numbers of slots for the schedule, as shown in Table 2.…”

Towards 48-fold cabling complexity reduction in large flattened butterfly networks

“…A drawback of these optical DC proposals is that they employ widely available optical switching devices, which switch in the order of milliseconds. However, micro-flow switching is prevalent in data center traffic patterns, i.e., microsecond flows are problematic to be handled with these available optical switching methods or current solutions do not scale well [26], [27], [28].…”

Dynamic optical switching for latency sensitive applications

“…Most algorithms in literature, such as TMS [18,33], Edmonds (used in Helios [17]) and similar others [11,38], require milliseconds to compute a feasible schedule for the interconnect. All of them share a key characteristic: a single centralized controller aggregates demands from all nodes before computing the circuit schedule.…”

“…Furthermore, as opposed to high complexity solutions such as the TMS, which has O(n 4.5 ) complexity [18], simpler demand collection and algorithms which can be easily implemented in hardware are preferred.…”