TY - GEN
T1 - Information Flow Queue Optimization in EC Cloud
AU - Tao, Yangyang
AU - Yu, Shucheng
AU - Zhou, Junxiu
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/6/19
Y1 - 2018/6/19
N2 - With the proliferation of big data applications, Erasure Codes (EC) have been increasingly adopted by distributed storage systems because of the space efficiency and high reliability. However, tail latency remains as a major limitation with EC which is responsible for long response times of many emerging Web applications. To combat tail latency of EC, one approach is to introduce real-time data displacement, which would cause additional latency due to data displacement itself. Queue models are frequently adopted for data scheduling. However, current queue model approaches usually overlook the real-time properties of requests and hence only provide suboptimal solutions. In this paper we consider the natures of real-time requests and the bipartite property of the network graph. We formulate the problem as a k-marriage flow queue model and optimize the scheduling strategy according to user preference (Service Level Agreement (SLA)) with multi-object optimization (MOO). Comprehensive simulation results show admissible improvement of the proposed method on tail latency as compared to the state of the art including the data displacement approach.
AB - With the proliferation of big data applications, Erasure Codes (EC) have been increasingly adopted by distributed storage systems because of the space efficiency and high reliability. However, tail latency remains as a major limitation with EC which is responsible for long response times of many emerging Web applications. To combat tail latency of EC, one approach is to introduce real-time data displacement, which would cause additional latency due to data displacement itself. Queue models are frequently adopted for data scheduling. However, current queue model approaches usually overlook the real-time properties of requests and hence only provide suboptimal solutions. In this paper we consider the natures of real-time requests and the bipartite property of the network graph. We formulate the problem as a k-marriage flow queue model and optimize the scheduling strategy according to user preference (Service Level Agreement (SLA)) with multi-object optimization (MOO). Comprehensive simulation results show admissible improvement of the proposed method on tail latency as compared to the state of the art including the data displacement approach.
KW - Erasure Codes
KW - Information Flow
KW - Optimization
KW - Queue theory
UR - http://www.scopus.com/inward/record.url?scp=85050133283&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85050133283&partnerID=8YFLogxK
U2 - 10.1109/ICCNC.2018.8390367
DO - 10.1109/ICCNC.2018.8390367
M3 - Conference contribution
AN - SCOPUS:85050133283
T3 - 2018 International Conference on Computing, Networking and Communications, ICNC 2018
SP - 888
EP - 892
BT - 2018 International Conference on Computing, Networking and Communications, ICNC 2018
T2 - 2018 International Conference on Computing, Networking and Communications, ICNC 2018
Y2 - 5 March 2018 through 8 March 2018
ER -