MPTCP is not pareto-optimal: performance issues and a possible solution
read more
Citations
Presto: Edge-based Load Balancing for Fast Datacenter Networks
Multipath TCP: analysis, design, and implementation
WiFi, LTE, or Both?: Measuring Multi-Homed Wireless Internet Performance
Reducing Internet Latency: A Survey of Techniques and Their Merits
Multipath Transmission for the Internet: A Survey
References
Congestion avoidance and control
Rate control for communication networks: shadow prices, proportional fairness and stability
The click modular router
TCP Congestion Control
CUBIC: a new TCP-friendly high-speed TCP variant
Related Papers (5)
Frequently Asked Questions (13)
Q2. What have the authors stated for future works in "Mptcp is not pareto-optimal: performance issues and a possible solution" ?
The stability and convergence of OLIA is another important question that will be studied in future work.
Q3. What is the way to measure the path conditions?
In [13], an opportunistic multipath scheduler measures the path conditions on time scales up to several seconds. [14] uses a mechanism to detect shared bottlenecks and to avoid the use of multiple subflows on the same bottleneck. [15] proposes to use uncoupled TCP flows with a weight depending on the congestion level.
Q4. What is the effect of type1 users on the throughput of type2?
As the number of type1 users increases, the throughput of type2 users decreases, but the throughput of type1 users does not change as it is limited by the capacity C1 of the streaming server.
Q5. How many times does OLIA increase p2?
In particular, the authors observe that by increasing N1 from 0 to 3N2, p2 increases by a factor of 2 using OLIA, whereas the increase is in the order of 4 to 6 times when using LIA.
Q6. What is the way to deal with a differential equation with a discontinuous righthand?
A natural way to deal with a differential equation with a discontinuous righthand size is to replace the differential equation (7) by a differential inclusion dx/dt ∈ F (x) where the discontinuous αr of (7) is replaced by the convex closure of the possible values of α in a small neighborhood of x [21, 22].
Q7. How does OLIA use the available capacity?
Their results show that although OLIA uses the available capacity as efficiently as LIA, the average completion time of short flows decreases by 10% using OLIA.
Q8. What is the first step to act on the minimum probing traffic rate?
A first one would be to act on the minimum probing traffic rate by an adjustment of the retransmit timer – in their current implementation, the minimum window size is 1 and the minimum probing rate is 1/rttr.
Q9. How much probing cost is needed to achieve the optimal traffic flow?
In this paper, the authors introduce a theoretical baseline for window-based congestion-control algorithms, called theoretical optimum with probing cost ; it provides optimal resource pooling in the network, given that a minimum probing traffic of 1 MSS per RTT is sent over each path.
Q10. What is the reason for the reduction in the aggregate throughput?
This 3.5% reduction in the aggregate throughput is due to the minimum traffic transmitted by users over congested paths and cannot be reduced as it is bounded below by 1/rtt packets/sec.6.1.3 Scenario C
Q11. How does OLIA solve the problem of LIA?
the authors have shown through measurements and by simulation that OLIA is as responsive and non-flappy as LIA, and that it solves identified problems with LIA.
Q12. what is the tcp u can use to determine the loss probability?
Ru wp(t)} (3)B(t) = { j(t) | j(t) = arg maxp∈Ru`p(t)rttp(t)2} (4)M(t) is the set of the paths of u with the largest window sizes at time t. B(t) is the set of the paths at time t that are presumably the best paths for u, as 1/`r(t) can be considered as an estimate of packet loss probability on path r at time t, and the rate that path r can provide to a TCP user can be estimated by √ 2`r(t)/rttr [17].
Q13. How much decrement in aggregate throughput when the authors update Red users to OLIA?
Their results show that there is a 3.5% decrement in aggregate throughput when the authors update Red users to OLIA, which is much smaller than the 13% reduction the authors observed when the authors used LIA (see Table 1).