Cost / performance trade-offs and fairness evaluation of queue mapping policies

Authors:
Teresa Nachiondo;José Flich;José Duato;Mitchell Gusat
Affiliations:
Dept. of Computer Engineering, Universidad Politécnica de Valencia, Valencia, Spain;Dept. of Computer Engineering, Universidad Politécnica de Valencia, Valencia, Spain;Dept. of Computer Engineering, Universidad Politécnica de Valencia, Valencia, Spain;IBM, Research, Zurich Research Lab. GmbH, Rueschlikon, Switzerland
Venue:
Euro-Par'05 Proceedings of the 11th international Euro-Par conference on Parallel Processing
Year:
2005

Citing 5
Cited 1

Virtual-channel flow control

ISCA '90 Proceedings of the 17th annual international symposium on Computer Architecture
Dynamic Voltage Scaling with Links for Power Optimization of Interconnection Networks

HPCA '03 Proceedings of the 9th International Symposium on High-Performance Computer Architecture
Principles and Practices of Interconnection Networks

Principles and Practices of Interconnection Networks
A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistage Interconnection Networks

HPCA '05 Proceedings of the 11th International Symposium on High-Performance Computer Architecture
Efficient Reduction of HOL Blocking in Multistage Networks

IPDPS '05 Proceedings of the 19th IEEE International Parallel and Distributed Processing Symposium (IPDPS'05) - Workshop 9 - Volume 10

Destination-Based HoL Blocking Elimination

ICPADS '06 Proceedings of the 12th International Conference on Parallel and Distributed Systems - Volume 1

Quantified Score

Hi-index	0.01

Visualization

Abstract

Whereas the established interconnection networks (ICTN) achieve low latency by operating in the linear region, i.e. oversizing the fabric, the recent strict cost and power constrains demand more efficient utilization of future networks. Increasing the utilization of lossless ICTNs may, however, lead to saturation and performance degradation owing to HOL-blocking. The current solution to HOL-blocking consists of using Virtual Output Queueing (VOQ), whose quadratical scalability is expensive in large networks. To improve VOQ's scalability we have proposed the Destination-Based Buffer Management (DBBM), a scheme that compares well with VOQ. Whereas previously we have analyzed DBBM's basic operation and performance, in this paper we have set two different goals. First we focus on how the different DBBM mappings can impact the cost/performance of multistage ICTNs. Next, because DBBM can introduce unfairness, this constitutes the second theme of our paper. The new results show that DBBM with modulo-4/8 mapping performs very well for only a fraction of the VOQ cost. Also in terms of fairness DBBM shows promise, because it (i) keeps the unfairness degree independent of both topology and routing, while (ii) minimizing the number of flows affected by unfairness.