Experimental queueing analysis with long-range dependent packet traffic
IEEE/ACM Transactions on Networking (TON)
ElephantTrap: A low cost device for identifying large flows
HOTI '07 Proceedings of the 15th Annual IEEE Symposium on High-Performance Interconnects
Measurement and analysis of TCP throughput collapse in cluster-based storage systems
FAST'08 Proceedings of the 6th USENIX Conference on File and Storage Technologies
Every microsecond counts: tracking fine-grain latencies with a lossy difference aggregator
Proceedings of the ACM SIGCOMM 2009 conference on Data communication
Safe and effective fine-grained TCP retransmissions for datacenter communication
Proceedings of the ACM SIGCOMM 2009 conference on Data communication
Understanding TCP incast throughput collapse in datacenter networks
Proceedings of the 1st ACM workshop on Research on enterprise networking
Proceedings of the ACM SIGCOMM 2010 conference
TritonSort: a balanced large-scale sorting system
Proceedings of the 8th USENIX conference on Networked systems design and implementation
Resource/accuracy tradeoffs in software-defined measurement
Proceedings of the second ACM SIGCOMM workshop on Hot topics in software defined networking
Hi-index | 0.00 |
The need to identify correlated traffic bursts at various, and especially fine-grain, time scales has become pressing in modern data centers. The combination of Gigabit link speeds and small switch buffers have led to "microbursts", which cause packet drops and large increases in latency. Our paper describes the design and implementation of an efficient and flexible end-host bandwidth measurement tool that can identify such bursts in addition to providing a number of other features. Managers can query the tool for bandwidth measurements at resolutions chosen after the traffic was measured. The algorithmic challenge is to support such a posteriori queries without retaining the entire trace or keeping state for all time scales. We introduce two aggregation algorithms, Dynamic Bucket Merge (DBM) and Exponential Bucketing (EXPB). We show experimentally that DBM and EXPB implementations in the Linux kernel introduce minimal overhead on applications running at 10 Gbps, consume orders of magnitude less memory than event logging (hundreds of bytes per second versus Megabytes per second), but still provide good accuracy for bandwidth measures at any time scale. Our techniques can be implemented in routers and generalized to detect spikes in the usage of any resource at fine time scales.