Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117589
Walid A. Y. Aljoby, T. Fu, Richard T. B. Ma
We consider data intensive cloud-based stream analytics where data transmission through the underlying communication network is the cause of the performance bottleneck. Two key inter-related problems are investigated: task placement and bandwidth allocation. We seek to answer the following questions. How does task placement make impact on the application-level throughput? Does a careful bandwidth allocation among data flows traversing a bottleneck link results in better performance? In this paper, we address these questions by conducting measurement-driven analysis in a SDN-enabled computer cluster running stream processing applications on top of Apache Storm. The results reveal (i) how tasks are assigned to computing nodes make large difference in application level performance; (ii) under certain task placement, a proper bandwidth allocation helps further improve the performance as compared to the default TCP mechanism; and (iii) task placement and bandwidth allocation are collaboratively making effects in overall performance.
{"title":"Impacts of task placement and bandwidth allocation on stream analytics","authors":"Walid A. Y. Aljoby, T. Fu, Richard T. B. Ma","doi":"10.1109/ICNP.2017.8117589","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117589","url":null,"abstract":"We consider data intensive cloud-based stream analytics where data transmission through the underlying communication network is the cause of the performance bottleneck. Two key inter-related problems are investigated: task placement and bandwidth allocation. We seek to answer the following questions. How does task placement make impact on the application-level throughput? Does a careful bandwidth allocation among data flows traversing a bottleneck link results in better performance? In this paper, we address these questions by conducting measurement-driven analysis in a SDN-enabled computer cluster running stream processing applications on top of Apache Storm. The results reveal (i) how tasks are assigned to computing nodes make large difference in application level performance; (ii) under certain task placement, a proper bandwidth allocation helps further improve the performance as compared to the default TCP mechanism; and (iii) task placement and bandwidth allocation are collaboratively making effects in overall performance.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"17 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89576128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117587
Yangming Zhao, Shouxi Luo, Yi Wang, Sheng Wang
Computing frameworks have been widely deployed to support global-scale services. A job typically has multiple sequential stages, where each stage is further divided into multiple parallel tasks. We call the set of all the tasks in a stage of a job a cotask. In this paper, we aim to minimize the average Cotask Completion Time (CCT) in cotask scheduling. To the best of our knowledge, there is no prior work on cotask scheduling for cloud computing. We propose the Cotask Scheduling Scheme (CSS), and take MapReduce as a representative of computing frameworks. CSS schedules cotasks following the Minimum Completion Time First (MCTF) policy, and we prove this problem is NP-hard. We formulate the model using the Integer Linear Programming (ILP), and solve it through an efficient heuristics based on ILP relaxation. Through real trace based simulations, we show that CSS is able to reduce the average CCT by up to 62.20% and 69.93% with traces from our testbed and from a large production cluster respectively.
计算框架已被广泛部署以支持全球规模的服务。一个作业通常有多个连续的阶段,其中每个阶段又进一步划分为多个并行任务。我们把作业的一个阶段中所有任务的集合称为协同任务。在本文中,我们的目标是最小化协同任务调度中的平均协同任务完成时间(CCT)。据我们所知,目前还没有关于云计算协同任务调度的研究。提出了协同任务调度方案(CSS),并以MapReduce为代表的计算框架。CSS调度协同任务遵循最小完成时间优先(Minimum Completion Time First, MCTF)策略,我们证明了这个问题是np困难的。我们使用整数线性规划(ILP)来建立模型,并通过基于ILP松弛的高效启发式求解。通过基于真实轨迹的模拟,我们表明CSS能够将平均CCT分别降低62.20%和69.93%,分别来自我们的测试平台和大型生产集群的轨迹。
{"title":"Cotask scheduling in cloud computing","authors":"Yangming Zhao, Shouxi Luo, Yi Wang, Sheng Wang","doi":"10.1109/ICNP.2017.8117587","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117587","url":null,"abstract":"Computing frameworks have been widely deployed to support global-scale services. A job typically has multiple sequential stages, where each stage is further divided into multiple parallel tasks. We call the set of all the tasks in a stage of a job a cotask. In this paper, we aim to minimize the average Cotask Completion Time (CCT) in cotask scheduling. To the best of our knowledge, there is no prior work on cotask scheduling for cloud computing. We propose the Cotask Scheduling Scheme (CSS), and take MapReduce as a representative of computing frameworks. CSS schedules cotasks following the Minimum Completion Time First (MCTF) policy, and we prove this problem is NP-hard. We formulate the model using the Integer Linear Programming (ILP), and solve it through an efficient heuristics based on ILP relaxation. Through real trace based simulations, we show that CSS is able to reduce the average CCT by up to 62.20% and 69.93% with traces from our testbed and from a large production cluster respectively.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"68 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79090659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117536
Yi Yang, G. Cao
Smartwatches are quickly gaining popularity, but their limited battery life remains an important factor that adversely affects user satisfaction. To provide full functionality, smartwatches are usually connected to phones via Bluetooth. However, the Bluetooth power characteristics and the energy impact of Bluetooth data traffic have been rarely studied. To address this issue, we first establish the Bluetooth power model based on extensive measurements and a thorough examination of the Bluetooth implementation on Android smartwatches. Then we perform the first in-depth investigation of the background data transfers on smartwatches, and find that they are prevalent and consume a large amount of energy. For example, our experiments show that the smartwatch's battery life can be reduced to one third (or even worse) due to background data transfers. Such high energy cost is caused by many unnecessary data transfers and the energy inefficiency attributed to the adverse interaction between the data transfer pattern (i.e., frequently transferring small data) and the Bluetooth energy characteristics (i.e., the tail effect). Based on the identified causes, we propose four energy optimization techniques, which are fast dormancy, phone-initiated polling, two-stage sensor processing, and context-aware pushing. The first one aims to reduce tail energy for delay-tolerant data transfers. The latter three are designed for specific applications which are responsible for most background data transfers. Evaluation results show that jointly using these techniques can save 70.6% of the Bluetooth energy.
{"title":"Characterizing and optimizing background data transfers on smartwatches","authors":"Yi Yang, G. Cao","doi":"10.1109/ICNP.2017.8117536","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117536","url":null,"abstract":"Smartwatches are quickly gaining popularity, but their limited battery life remains an important factor that adversely affects user satisfaction. To provide full functionality, smartwatches are usually connected to phones via Bluetooth. However, the Bluetooth power characteristics and the energy impact of Bluetooth data traffic have been rarely studied. To address this issue, we first establish the Bluetooth power model based on extensive measurements and a thorough examination of the Bluetooth implementation on Android smartwatches. Then we perform the first in-depth investigation of the background data transfers on smartwatches, and find that they are prevalent and consume a large amount of energy. For example, our experiments show that the smartwatch's battery life can be reduced to one third (or even worse) due to background data transfers. Such high energy cost is caused by many unnecessary data transfers and the energy inefficiency attributed to the adverse interaction between the data transfer pattern (i.e., frequently transferring small data) and the Bluetooth energy characteristics (i.e., the tail effect). Based on the identified causes, we propose four energy optimization techniques, which are fast dormancy, phone-initiated polling, two-stage sensor processing, and context-aware pushing. The first one aims to reduce tail energy for delay-tolerant data transfers. The latter three are designed for specific applications which are responsible for most background data transfers. Evaluation results show that jointly using these techniques can save 70.6% of the Bluetooth energy.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"25 1","pages":"1-10"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78075026","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117598
T. Kunz, K. Muthukumar
SDN technology has been applied to a range of different networks, ranging from Ethernet services to large cloud environments. More recently, interest has turned towards extending programmability of Optical Transport Networks (OTN). In the SDN architecture, SBIs are used to communicate between the SDN controller and the switches or routers in the network. In this paper, we deploy OpenFlow and NETCONF as SBIs in managing BoD across interconnected data centers over OTN. More specifically, we use these protocols to communicate between an OpenDayLight controller and two BTI7800 network elements that interconnect the data centers. We present experimental results for both BTI's YANG-based NETCONF implementation and our port of OpenFlow for a number of use cases. Our results show that NETCONF is faster and requires fewer control message. However, OpenFlow offers better bandwidth utilization over the interconnecting link.
{"title":"Comparing OpenFlow and NETCONF when interconnecting data centers","authors":"T. Kunz, K. Muthukumar","doi":"10.1109/ICNP.2017.8117598","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117598","url":null,"abstract":"SDN technology has been applied to a range of different networks, ranging from Ethernet services to large cloud environments. More recently, interest has turned towards extending programmability of Optical Transport Networks (OTN). In the SDN architecture, SBIs are used to communicate between the SDN controller and the switches or routers in the network. In this paper, we deploy OpenFlow and NETCONF as SBIs in managing BoD across interconnected data centers over OTN. More specifically, we use these protocols to communicate between an OpenDayLight controller and two BTI7800 network elements that interconnect the data centers. We present experimental results for both BTI's YANG-based NETCONF implementation and our port of OpenFlow for a number of use cases. Our results show that NETCONF is faster and requires fewer control message. However, OpenFlow offers better bandwidth utilization over the interconnecting link.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"47 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73908859","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117588
Dagang Li, Zehua Zheng
Disk-based large scale graph computation on a single machine has been attracting much attention, with GraphChi as one of the most well-accepted solutions. However, we find out that the performance of GraphChi becomes I/O-constrained when memory is moderately abundant, and from some point adding more memory does not help with the performance any more. In this work, a greedy caching algorithm GSC is proposed for GraphChi to make better use of the memory. It alleviates the I/O constraint by caching and delaying the write-backs of GraphChi shards that have already been loaded into the memory. Experimental results show that by minimizing unnecessary I/Os, GSC can be up to 4x faster during computation than standard GraphChi under memory constraint, and achieve about 3x performance gain when sufficient memory is available.
在单个机器上基于磁盘的大规模图计算已经引起了很多关注,GraphChi是最被广泛接受的解决方案之一。然而,我们发现,当内存适度充裕时,GraphChi的性能会受到I/ o约束,从某种程度上说,增加更多的内存对性能不再有帮助。为了更好地利用GraphChi的内存,本文提出了一种贪心缓存算法GSC。它通过缓存和延迟已经加载到内存中的GraphChi分片的回写来缓解I/O约束。实验结果表明,通过最小化不必要的I/ o, GSC在内存限制下的计算速度可以比标准GraphChi快4倍,并且在足够的内存可用时可以实现约3倍的性能提升。
{"title":"GSC: Greedy shard caching algorithm for improved I/O efficiency in GraphChi","authors":"Dagang Li, Zehua Zheng","doi":"10.1109/ICNP.2017.8117588","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117588","url":null,"abstract":"Disk-based large scale graph computation on a single machine has been attracting much attention, with GraphChi as one of the most well-accepted solutions. However, we find out that the performance of GraphChi becomes I/O-constrained when memory is moderately abundant, and from some point adding more memory does not help with the performance any more. In this work, a greedy caching algorithm GSC is proposed for GraphChi to make better use of the memory. It alleviates the I/O constraint by caching and delaying the write-backs of GraphChi shards that have already been loaded into the memory. Experimental results show that by minimizing unnecessary I/Os, GSC can be up to 4x faster during computation than standard GraphChi under memory constraint, and achieve about 3x performance gain when sufficient memory is available.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"4 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86764390","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117564
Zhuotao Liu, Yushan Liu, Philipp Winter, Prateek Mittal, Yih-Chun Hu
Tor is the most widely used anonymity network, currently serving millions of users each day. However, there is no access control in place for all these users, leaving the network vulnerable to botnet abuse and attacks. For example, criminals frequently use exit relays as stepping stones for attacks, causing service providers to serve CAPTCHAs to exit relay IP addresses or blacklisting them altogether, which leads to severe usability issues for legitimate Tor users. To address this problem, we propose TorPolice, the first privacy-preserving access control framework for Tor. TorPolice enables abuse-plagued service providers such as Yelp to enforce access rules to police and throttle malicious requests coming from Tor while still providing service to legitimate Tor users. Further, TorPolice equips Tor with global access control for relays, enhancing Tor's resilience to botnet abuse. We show that TorPolice preserves the privacy of Tor users, implement a prototype of TorPolice, and perform extensive evaluations to validate our design goals.
{"title":"TorPolice: Towards enforcing service-defined access policies for anonymous communication in the Tor network","authors":"Zhuotao Liu, Yushan Liu, Philipp Winter, Prateek Mittal, Yih-Chun Hu","doi":"10.1109/ICNP.2017.8117564","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117564","url":null,"abstract":"Tor is the most widely used anonymity network, currently serving millions of users each day. However, there is no access control in place for all these users, leaving the network vulnerable to botnet abuse and attacks. For example, criminals frequently use exit relays as stepping stones for attacks, causing service providers to serve CAPTCHAs to exit relay IP addresses or blacklisting them altogether, which leads to severe usability issues for legitimate Tor users. To address this problem, we propose TorPolice, the first privacy-preserving access control framework for Tor. TorPolice enables abuse-plagued service providers such as Yelp to enforce access rules to police and throttle malicious requests coming from Tor while still providing service to legitimate Tor users. Further, TorPolice equips Tor with global access control for relays, enhancing Tor's resilience to botnet abuse. We show that TorPolice preserves the privacy of Tor users, implement a prototype of TorPolice, and perform extensive evaluations to validate our design goals.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"41 1","pages":"1-10"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90844462","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117538
Yingjie Bi, A. Tang
This paper studies routing performance loss due to traffic split ratio granularity constraints. For any given fineness of traffic splitting capability, we provide upper bounds on the loss of performance. Based on tight convex relaxation, we further develop an efficient approximation algorithm to compute a good routing solution that satisfies given constraints on traffic splitting. The results can be useful for network operators to trade-off between attainable performance and implementation overhead. Some of the mathematical techniques developed here can be of independent interest for studying other similar nonconvex optimization problems.
{"title":"Cost of not arbitrarily splitting in routing","authors":"Yingjie Bi, A. Tang","doi":"10.1109/ICNP.2017.8117538","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117538","url":null,"abstract":"This paper studies routing performance loss due to traffic split ratio granularity constraints. For any given fineness of traffic splitting capability, we provide upper bounds on the loss of performance. Based on tight convex relaxation, we further develop an efficient approximation algorithm to compute a good routing solution that satisfies given constraints on traffic splitting. The results can be useful for network operators to trade-off between attainable performance and implementation overhead. Some of the mathematical techniques developed here can be of independent interest for studying other similar nonconvex optimization problems.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"12 1","pages":"1-10"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81921824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117580
Max Engelhardt, A. Asadi
Experimental setups for cellular communications have always been a rare commodity in academia. The advent of software-defined radios (SDRs) paved the way for researchers to prototype their ideas on real hardware. However, existing SDR platforms and their associated reference design codes mostly provide basic cellular functionality with limitations such as low numbers of users and computational capacity. In this demo, we demonstrate the first SDR-based testbed for inband D2D communications using LabVIEW Communications and the USRP hardware platform. Furthermore, we implement a light-weight quality-aware scheduler which adaptively switches communication links from D2D to cellular and vice versa.
{"title":"The first experimental SDR platform for inband D2D communications in 5G","authors":"Max Engelhardt, A. Asadi","doi":"10.1109/ICNP.2017.8117580","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117580","url":null,"abstract":"Experimental setups for cellular communications have always been a rare commodity in academia. The advent of software-defined radios (SDRs) paved the way for researchers to prototype their ideas on real hardware. However, existing SDR platforms and their associated reference design codes mostly provide basic cellular functionality with limitations such as low numbers of users and computational capacity. In this demo, we demonstrate the first SDR-based testbed for inband D2D communications using LabVIEW Communications and the USRP hardware platform. Furthermore, we implement a light-weight quality-aware scheduler which adaptively switches communication links from D2D to cellular and vice versa.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"22 1","pages":"1-2"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88635448","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The reactive model of Software Defined Networking (SDN) invokes controller to dynamically determine the behaviors of a new flow without any pre-knowledge in the data plane. However, the reactive events raised by such flexible model meanwhile consume lots of the bottleneck resources of the fast memory in switch and bandwidth between controller and switches. To address this problem, we propose SoftRing with the motivation to mitigate the overhead to handle a reactive event. In fact, the reactive packets are not necessarily stored in the switch or sent to the controller; instead, they are forwarded to traverse a pre-defined loop path. The packets will finally leave the loop path after the switch rules related to the packet flow being updated to switches in the loop with fewer flow entries. We have implemented a SoftRing system that integrates the controller and software/hardware SDN switches. The results show that SoftRing can eliminate the fast memory requirement for reactive packets and reduce the control channel bandwidth consumption up to 80%, with the cost of less than 5% data plane bandwidth, an average of three extra flow entries in each switch, and minor extra latency for the flow forwarding.
{"title":"SoftRing: Taming the reactive model for software defined networks","authors":"Chengchen Hu, Kaiyu Hou, Hao Li, Ruilong Wang, Peng Zheng, Peng Zhang, Huanzhao Wang","doi":"10.1109/ICNP.2017.8117558","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117558","url":null,"abstract":"The reactive model of Software Defined Networking (SDN) invokes controller to dynamically determine the behaviors of a new flow without any pre-knowledge in the data plane. However, the reactive events raised by such flexible model meanwhile consume lots of the bottleneck resources of the fast memory in switch and bandwidth between controller and switches. To address this problem, we propose SoftRing with the motivation to mitigate the overhead to handle a reactive event. In fact, the reactive packets are not necessarily stored in the switch or sent to the controller; instead, they are forwarded to traverse a pre-defined loop path. The packets will finally leave the loop path after the switch rules related to the packet flow being updated to switches in the loop with fewer flow entries. We have implemented a SoftRing system that integrates the controller and software/hardware SDN switches. The results show that SoftRing can eliminate the fast memory requirement for reactive packets and reduce the control channel bandwidth consumption up to 80%, with the cost of less than 5% data plane bandwidth, an average of three extra flow entries in each switch, and minor extra latency for the flow forwarding.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"49 1","pages":"1-10"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74926682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-10-01DOI: 10.1109/ICNP.2017.8117545
F. Duchene, O. Bonaventure
Multipath TCP is a recent TCP extension that enables the utilization of different paths for a single connection. This provides various benefits including bandwidth aggregation and fast handovers on mobiles. A Multipath TCP connection starts with a single TCP connection called subflow and other subflows are added later to increase bandwidth or support failover. One drawback of Multipath TCP is that it is not currently compatible with stateless load balancers which rely on the five-tuple for their forwarding decision. This hinders the deployment of Multipath TCP. We show that this limitation can be circumvented with a small change to the handling of the initial subflow. Clients use this subflow to discover the load-balanced server and the additional Multipath TCP subflows are terminated at a unique address associated to each physical server. With this small change, Multipath TCP becomes compatible with existing stateless load balancers. Furthermore, we show that the same approach enables anycast Multipath TCP services, a major benefit given the difficulty of deploying anycast TCP services. We implement this modification in the Linux kernel and demonstrate its benefits with several micro benchmarks.
{"title":"Making multipath TCP friendlier to load balancers and anycast","authors":"F. Duchene, O. Bonaventure","doi":"10.1109/ICNP.2017.8117545","DOIUrl":"https://doi.org/10.1109/ICNP.2017.8117545","url":null,"abstract":"Multipath TCP is a recent TCP extension that enables the utilization of different paths for a single connection. This provides various benefits including bandwidth aggregation and fast handovers on mobiles. A Multipath TCP connection starts with a single TCP connection called subflow and other subflows are added later to increase bandwidth or support failover. One drawback of Multipath TCP is that it is not currently compatible with stateless load balancers which rely on the five-tuple for their forwarding decision. This hinders the deployment of Multipath TCP. We show that this limitation can be circumvented with a small change to the handling of the initial subflow. Clients use this subflow to discover the load-balanced server and the additional Multipath TCP subflows are terminated at a unique address associated to each physical server. With this small change, Multipath TCP becomes compatible with existing stateless load balancers. Furthermore, we show that the same approach enables anycast Multipath TCP services, a major benefit given the difficulty of deploying anycast TCP services. We implement this modification in the Linux kernel and demonstrate its benefits with several micro benchmarks.","PeriodicalId":6462,"journal":{"name":"2017 IEEE 25th International Conference on Network Protocols (ICNP)","volume":"18 1","pages":"1-10"},"PeriodicalIF":0.0,"publicationDate":"2017-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87033723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}