Chengchen Hu

Survey on routing in data centers: insights and future directions

Recently, a series of data center network architectures have been proposed. The goal of these works is to interconnect a large number of servers with significant bandwidth requirements. Coupled with these new DCN structures, routing protocols play an important role in exploring the network capacities that can be potentially delivered by the topologies. This article conducts a survey on the current state of the art of DCN routing techniques. The article focuses on the insights behind these routing schemes and also points out the open research issues hoping to spark new interests and developments in this field.

A TCAM-based distributed parallel IP lookup scheme and performance analysis

Using ternary content addressable memory (TCAM) for high-speed IP address lookup has been gaining popularity due to its deterministic high performance. However, restricted by the slow improvement of memory accessing speed, the route lookup engines for next-generation terabit routers demand exploiting parallelism among multiple TCAM chips. Traditional parallel methods always incur excessive redundancy and high power consumption. We propose in this paper an original TCAM-based IP lookup scheme that achieves both ultra-high lookup throughput and optimal utilization of the memory while being power-efficient. In our multi-chip scheme, we devise a load-balanced TCAM table construction algorithm together with an adaptive load balancing mechanism. The power efficiency is well controlled by decreasing the number of TCAM entries triggered in each lookup operation. Using four 133 MHz TCAM chips and given 25% more TCAM entries than the original route table, the proposed scheme achieves a lookup throughput of up to 533 MPPS while remains simple for ASIC implementation

Towards a secure controller platform for openflow applications

The OpenFlow (OF) paradigm embraces third-party development efforts, and therefore suffers from potential trust issue on OF applications (apps). The abuse of such trust could lead to various types of attacks impacting the entire network. In this paper, we propose PermOF, a fine-grained permission system, as the first line of defense, in order to apply minimum privilege on apps. We summarize a set of 18 permissions to be enforced at the API entry of the controller. To accommodate the isolation requirements, we propose a customized isolation mechanism, which achieves comprehensive resource isolation and access control.

An ultra high throughput and power efficient TCAM-based IP lookup engine

Ternary content-addressable memory (TCAM) is widely used in high-speed route lookup engines. However, restricted by the memory access speed, the route lookup engines for next-generation terabit routers demand exploiting parallelism among multiple TCAMs. Traditional parallel methods always incur excessive redundancy and high power consumption. We propose in this paper an original TCAM-based IP lookup scheme that achieves an ultra high lookup throughput and a high utilization of the memory while being power efficient. In our multichip scheme, we devise a load-balanced TCAM table construction algorithm together with an adaptive load balancing mechanism. The power efficiency is well controlled by decreasing the number of TCAM entries triggered in each lookup operation. Using 133 MHz TCAM chips and given 25% more TCAM entries than the original route table, the proposed scheme achieves a lookup throughput of up to 533 Mpps and is simple for ASIC implementation

Accurate and Efficient Traffic Monitoring Using Adaptive Non-Linear Sampling Method

Sampling technology has been widely deployed in measurement systems to control memory consumption and processing overhead. However, most of the existing sampling methods suffer from large estimation errors in analyzing small-size flows. To address the problem, we propose a novel adaptive non-linear sampling (ANLS) method for passive measurement. Instead of statically configuring the sampling rate, ANLS dynamically adjusts the sampling rate for a flow depending on the number of packets having been counted. We provide the generic principles guiding the selection of sampling function for sampling rate adjustment. Moreover, we derive the unbiased flow size estimation, the bound of the relative error, and the bound of required counter size for ANLS. The performance of ANLS is thoroughly studied through theoretic analysis and experiments under synthetic/real network data traces, with comparison to several related sampling methods. The results demonstrate that the proposed ANLS can significantly improve the estimation accuracy, particularly for small-size flows, while maintain a memory and processing overhead comparable to existing methods.

Route Table Partitioning and Load Balancing for Parallel Searching with TCAMs

With the continuous advances in optical communications technology, the link transmission speed of Internet backbone has been increasing rapidly. This in turn demands more powerful IP address lookup engine. In this paper, we propose a power-efficient parallel TCAM-based lookup engine with a distributed logical caching scheme for dynamic load-balancing. In order to distribute the lookup requests among multiple TCAM chips, a smart partitioning approach called pre-order splitting divides the route table into multiple sub-tables for parallel processing. Meanwhile, by virtual of the cache-based load balancing scheme with slow-update mechanism, a speedup factor ofN-1 can be guaranteed for a system with N (N>2) TCAM chips, even with unbalanced bursty lookup requests.

WaveCube: A scalable, fault-tolerant, high-performance optical data center architecture

Optical data center networks (DCNs) are becoming increasingly attractive due to their technological strengths compared to traditional electrical networks. However, prior optical DCNs are either hard to scale, vulnerable to single point of failure, or provide limited network bisection bandwidth for many practical DCN workloads. To this end, we present WaveCube, a scalable, fault-tolerant, high-performance optical DCN architecture. To scale, WaveCube removes MEMS1, a potential bottleneck, from its design. Wave-Cube is fault-tolerant since it does not have single point of failure and there are multiple node-disjoint parallel paths between any pair of Top-of-Rack (ToR) switches. WaveCube delivers high performance by exploiting multi-pathing and dynamic link bandwidth along the path. Our extensive evaluation results show that WaveCube outperforms previous optical DCNs by up to 400% and delivers network bisection bandwidth that is 70%-85% of an ideal non-blocking network under both realistic and synthetic traffic patterns. WaveCubes performance degrades gracefully under failures - it drops 20% even with 20% links cut. WaveCube also holds promise in practice - its wiring complexity is orders of magnitude lower than Fattree, BCube and c-Through at large scale, and its power consumption is 35% of them.

VirtualKnotter: Online virtual machine shuffling for congestion resolving in virtualized datacenter

Abstract Our measurements on production datacenter traffic together with recently-reported results (Kandula et al.) [1] suggest that datacenter networks suffer from long-lived congestion caused by core network oversubscription and unbalanced workload placement. In contrast to traditional traffic engineering approaches that optimize flow routing, in this paper, we explore the opportunity to address the continuous congestion via optimizing VM placement in virtualized datacenters. To this end, we present VirtualKnotter to reduce congestion with controllable VM migration traffic as well as low migration time, which includes an online VM placement algorithm and an efficient VM migration scheduling algorithm. Our evaluation with both real and synthetic traffic patterns shows that VirtualKnotter performs close to the baseline algorithm in terms of link unitization, with only 5–10% migration traffic of the baseline algorithm. Furthermore, VirtualKnotter decreases link congestion time by 53% for the production datacenter traffic.

Per-Flow Queueing by Dynamic Queue Sharing

Per-flow queuing is believed to be able to guarantee advanced Quality of Service (QoS) for each flow. With the dramatic increase of link speed and number of traffic flows, per-flow queuing faces a great challenge since millions of queues need to be maintained for implementation in a traditional sense. In this paper, by setting only a small number of physical queues, we propose a Dynamic Queue Sharing (DQS) mechanism to achieve an equal performance to the pure per-flow queuing with a lower cost. The proposed mechanism is based on an interesting fact that the number of simultaneous active flows in the router buffer is far less than that of in-progress flows. In DQS, a physical queue is dynamically created on-demand when a new flow comes and then dynamically released when the flow temporarily pauses. Hashing and binary sorting tree (or linked list) are combined to manage the mapping between flows and queues, so as to isolate flows in different queues. Theoretical analysis and traces experiments are conducted to evaluate DQS. The results demonstrate that when the parameters are well set, the operation delay is less than two time cycles in average with an extra memory of 16k bits.

VirtualKnotter: Online Virtual Machine Shuffling for Congestion Resolving in Virtualized Datacenter

Our measurements on production data center traffic together with recently reported results suggest that data center networks suffer from long-lived congestion caused by core network over subscription and unbalanced workload placement. In contrast to traditional traffic engineering approaches that optimize flow routing, in this paper, we explore the opportunity to address the continuous congestion via optimizing VM placement in virtualized data centers. To this end, we present Virtual Knotter, an efficient online VM placement algorithm to reduce congestion with controllable VM migration traffic as well as low time complexity. Our evaluation with both real and synthetic traffic patterns shows that Virtual Knotter performs close to the baseline algorithm in terms of link unitization, with only 5%-10% migration traffic of the baseline algorithm. Furthermore, Virtual Knotter decreases link congestion time by 53% for the production data center traffic.