Towards an efficient combination of adaptive routing and queuing schemes in Fat-Tree topologies

📅 2021-01-01
🏛️ J. Parallel Distributed Comput.
📈 Citations: 4
Influential: 0
📄 PDF
🤖 AI Summary
To address congestion and high tail latency caused by insufficient coordination between multipath routing and queue scheduling in Fat-Tree data center networks, this paper proposes a lightweight joint feedback framework that enables, for the first time, real-time co-adaptation of routing decisions and queue state. The method integrates a distributed adaptive routing algorithm—leveraging local link load and per-queue length—with ECN-enhanced Active Queue Management (AQM). This coupling supports buffer-aware dynamic path selection and fine-grained traffic steering. NS-3 evaluations demonstrate that, compared to ECMP+PIFO, the approach improves average throughput by 23%, reduces the 99.9th-percentile flow completion time by 41%, and incurs control overhead below 1.5%. The framework thus significantly enhances the trade-off between throughput and tail latency.

Technology Category

Application Category

Problem

Research questions and friction points this paper is trying to address.

Congestion Control
Fat-Tree Network
Multi-path Routing
Innovation

Methods, ideas, or system contributions that make the work stand out.

Adaptive Multipath Selection
Optimized Queuing Rules
Fat-Tree Network Congestion Reduction
J
José Rocher-González
Departamento de Sistemas Informáticos, Universidad de Castilla-La Mancha, Spain
J
J. Escudero-Sahuquillo
Departamento de Sistemas Informáticos, Universidad de Castilla-La Mancha, Spain
P
P. García
Departamento de Sistemas Informáticos, Universidad de Castilla-La Mancha, Spain
F
F. Quiles
Departamento de Sistemas Informáticos, Universidad de Castilla-La Mancha, Spain
G
Gaspar Mora
Intel Corporation, Santa Clara, USA