負載平衡器在資料中心中已是極為常見的必需品。隨著雲端需求不斷增長，為了提供更大規模的單一服務，負載平衡器的重要性與日俱增。資料中心裡的網路服務依賴負載平衡器為其整合下層眾多的伺服器，整合為一個單一的服務。因此若一個服務想要快速反應，負載平衡器的效能不可成為瓶頸。傳統的負載平衡器非常昂貴，且其使用成本隨著流量增加而大增，因此一部份資料中心管理者開始尋求軟體負載平衡方案。負載平衡軟體表面上成本低廉，然而由於運行於一般電腦架構上，軟體負載平衡器並不適合大量處理封包，造成過高的CPU使用率，因此需添購額外電腦，造成成本上升。本篇方法 ECMP Static Route (ESR)為一種軟硬體綜合的負載平衡系統。ESR利用資料中心交換機的原有功能建構出硬體負載平衡器。交換機有能夠快速處理大量封包的優勢，且存在於每個資料中心中。利用交換機建構負載平衡器，將可以同時擁有高效能與低成本。使用ESR的風險很低，因為不需要購買額外硬體。我們在思科的Virtual Internet Routing Lab上試驗了本系統的可行性，並在omnet++上模擬了本篇方法的效能數據。

Service load balancers are critical components in modern data centers. Load balancers aim to split the traffic of services evenly among servers. Traditional load balancers are dedicated devices for load balancing, but the cost is unacceptable to many data centers. As a result, software load balancers are chosen as an alternative. However, software load balancers run on servers, and while cost less in licensing, running load balancing with CPU is inefficient resulting low throughput. Recently, Duet \cite{Gandhi2014} utilizes tunneling technique and switches in data centers as low-cost load balancers. However, DUET has the problem of inefficiency using non-optimal routes. Since in most data centers, the oversubscription ratio is not 1:1, it is of particular importance to avoid unnecessary bandwidth usage. We present ECMP Static Route (ESR) in this paper. ESR is a hybrid method combining switch functionalities and server software to form a load balancing system, having the benefit of high performance, optimal routing path, and low cost. We reuse the ECMP functionality in switches to provide load balancing. Our design provides a complete solution including load balancing and failure recovery. We implemented ESR in Cisco Virtual Internet Routing Lab to prove the feasibility of the idea. Also, we used OMNeT++ to evaluate the performance of the system. The evaluation shows that our method has 1/3 improvement in flow completion time.

[1] Rohan Gandhi, Hongqiang Harry Liu, Y. Charlie Hu, Guohan Lu, Jitendra Padhye, Lihua Yuan, and Ming Zhang. Duet: Cloud scale load balancing with hardware and software. In Proceedings of the 2014 ACM Conference on SIGCOMM, SIGCOMM ’14, pages 27–38, New York, NY, USA, 2014. ACM.
[2] F5 big-ip.
[3] Parveen Patel, Deepak Bansal, Lihua Yuan, Ashwin Murthy, Albert Greenberg, David A. Maltz, Randy Kern, Hemant Kumar, Marios Zikos, Hongyu Wu, Changhoon Kim, and Naveen Karri. Ananta: Cloud scale load balancing. In Proceedings of the ACM SIGCOMM 2013 Conference on SIGCOMM, SIGCOMM ’13, pages 207–218, New York, NY, USA, 2013. ACM.
[4] Don MacVittie. Intro to load balancing for developers – the algorithms, 6 2010.
[5] Willy Tarreau. HAProxy Configuration Manual, 3 2016.
[6] Richard Wang, Dana Butnariu, and Jennifer Rexford. Openflow-based server load balancing gone wild. In Proceedings of the 11th USENIX Conference on Hot Topics in Management of Internet, Cloud, and Enterprise Networks and Services, Hot-ICE’11, pages 12–12, Berkeley, CA, USA, 2011. USENIX Association.
[7] Rohan Gandhi, Y. Charlie Hu, Cheng kok Koh, Hongqiang (Harry) Liu, and Ming Zhang. Rubik: Unlocking the power of locality and end-point flexibility in cloud scale load balancing. In 2015 USENIX Annual Technical Conference (USENIX ATC 15), pages 473–485, Santa Clara, CA, July 2015. USENIX Association.
[8] Cisco intelligent traffic director.
[9] NanditaDukkipatiandNickMcKeown.Whyflow-completiontimeistherightmetric for congestion control. SIGCOMM Comput. Commun. Rev., 36(1):59–62, January 2006.