References in periodicals archive ?
The second target of OFPT is to improve the network throughput, i.e., improving the utilization [lambda] of a given bisection bandwidth.
OFPT aims to minimize the bandwidth utilization of each link in the network for the given traffic, which is equivalent to keeping the capacities of the link fixed but scale the injected traffic.
We use Fully Polynomial Time Approximation Scheme (FPTAS) to solve this problem, which is simple to implement and runs significantly faster than a general linear programming solver in OFPT.
2) OFPT maintains path information in the controller, which can provide all path information for a flow directly rather than compute everything from scratch.
We evaluate the performance of OFPT in Mininet  emulator.
4) OFPT: Our design is described in section 5, and we use TCP-New Reno in servers.
Metrics: We use throughput and fairness to evaluate the performance of OFPT. The throughput is calculated by [B.sub.flow]/Bisection Bandwidth, where [B.sub.flow] is the throughput of the flow, and Bisection Bandwidth is the available bandwidth provided by network.
We firstly test the performance of OFPT in one-to-one workload, and compare it with TCP, ECMP and MPTCP.
4(a) and 4(b) show the throughput of TCP, ECMP, MPTCP (the number of subflow is from 2 to 8, and we use 2 to represent the MPTCP with 2 subflows) and OFPT in k=8 FatTree and VL2.
4(c) and 4(d) show the bandwidth utilization of each flow with TCP, ECMP, MPTCP and OFPT in FatTree and VL2 respectively.
In conclusion, OFPT can efficiently improve the throughput and fairness in both FatTree and VL2 under one-to-one communication model.
In order to verify the performance of OFPT in different communication patterns, we test the performance of OFPT under different workload in this section.
Acronyms browser ?
Full browser ?
- Ofotbanen AS