Fine-tunning ECN and RTT based Congestion Control in Data Centers Networks Test phase, srs, design phase and source code final deliverable
Project Domain / Category
Networking
Abstract / Introduction
Web packages are progressively moving into cloud environments. These cloud offerings are hosted on big scale computation and garage infrastructures referred to as records centers (DC) (e.G. Google’s statistics center, Facebook’s records middle etc.). In a huge scale data center, loads of thousands of servers are interconnected through switches in a symmetric topology. A statistics center community (DCN) interconnects all of the facts middle resources.
Current information center congestion manage schemes may additionally induce excessive latency in packet shipping because of direction’s latency-oblivious congestion sign. Explicit congestion notification (ECN) is the predominantly used congestion signal in the facts middle networks; it indicators whether any queue along the direction is above a predefined threshold or no longer, but does no longer inform about the cease-to-stop postpone of the direction. Information of the give up-to-cease postpone / round-journey-time (RTT) of a course can enable a sender to regulate its sending charge to hold the community latency under a threshold. RTTs in statistics center networks are on the dimensions of as much as few hundreds of microseconds and traditional facts center running structures lack such pleasant-grained microsecond-degree timers; that is why postpone-primarily based congestion control schemes, which might be broadly deployed in the Internet, have not been used in the information center networks. But current studies like [1] advise that advances in NIC hardware has enabled correct microsecond-stage RTT size, thus, RTT can be integrated in the information center congestion manipulate schemes.
This undertaking targets to mix ECN and RTT indicators in the records middle congestion manipulate scheme and fine-track it. First of all, RTT size (on the source) could be carried out; then the prevailing ECN will be mixed with the RTT inside the congestion manipulate scheme; lastly the congestion manipulate algorithm (based at the ECN and RTT) may be first-rate-tunned to offer the most beneficial results. All implementation could be achieved in community simulator ns-2. The students will: i) observe an current broadly-deployed DCN congestion control scheme and its implementation in ns-2, ii) design adjustments inside the current set of rules(s) to contain put off/RTT, iii) implement the design inside the current congestion manage scheme in ns-2.
Working in ns-2 requires: i) information of fundamental instructions of Linux working systems (for ns-2 installation and walking functions), ii) remarkable programming skills in C++ (for simulating the DCN surroundings and implementing the proposed modifications in the existing congestion manage scheme), iii) programming in TCL (for writing simulation scripts), iv) knowledge of AWK command (for hint text processing), and v) expertise GNUPLOT command (for drawing graphs).
References:
[1] Mittal, R., Lam, V. T., Dukkipati, N., Blem, E., Wassel, H., Ghobadi, M., … & Zats, D. (2015). TIMELY: RTT-based totally Congestion Control for the Datacenter. ACM SIGCOMM Computer Communication Review, forty five(4), 537-550.
Intended Outcome
• Design of a direction’s latency-conscious DCN congestion control scheme
• Implementation of a path’s latency-conscious DCN congestion manipulate scheme in ns-2
Required Skills
1. Understanding of fundamental Linux instructions
2. C++, TCL, AWK and GNUPLOT (for running in ns-2)