Understanding congestion in high performance interconnection networks using sampling

Abstract

Understanding Congestion in High Performance Interconnection Networks Using Sampling by Philip A. Taffet The computational needs of many applications outstrip the capabilities of a single compute node. Communication is necessary to employ multiple nodes, but slow communication often limits application performance on multiple nodes. To improve communication performance, developers need tools that enable them to understand how their application's communication patterns interact with the network, especially when those interactions result in congestion. Since communication performanceis difficult to reason about analytically and simulation is costly, measurement-based approaches are needed. This thesis describes a new sampling-based technique to collect information about the path a packet takes and congestion it encounters. Experiments with simulations show that this strategy can distinguish problems with an application's communication patterns, its mapping onto a parallel system, and outside interference. We describe a variant of this scheme that requires only 5-6 bits of information in a monitored packet, making it practical for use in next-generation networks.

PINT: Probabilistic In-band Network Telemetry

“…However, they attempt to trace back to potential attackers (e.g., they do not assume unique packet IDs or reliable TTL values as these can be forged) and require significantly more packets for identification, as we show in Section 6. In a recent effort to reduce overheads on packets, similarly to this work, Taffet et al [71] propose having switches use Reservoir Sampling to collect information about a packet's path and congestion that the packet encounters as it passes through the network. PINT takes the process several steps further, including approximations and coding (XORbased or network coding) to reduce the cost of adding information to packets as much as possible.…”

An In-Depth Analysis of the Slingshot Interconnect

“…Regarding the evaluation of the under-load system, different works analyzed the impact of congestion (also known as network noise) on application performance [5], [6], [11], [71]- [74] on different types of networks. The GPCNet benchmark [6] has been recently proposed as a portable benchmark for estimating network congestion.…”

Bring orders into uncertainty

“…Some community efforts [14,28,38,45,53,54,59] have been made to efficiently process and analyze these data via exploiting highperformance accelerators under a heterogeneous scale-up and scaleout setup which has become mainstream node architecture for Top500 supercomputers. Among these important graph analytics, uncertainty is often intrinsic to a wide spectrum of graph applications, which applies to graph data such as noisy measurement in inter-node connection in supercomputing center [38,55], database querying [7,12,25,26,29], probability in peer-to-peer network [25], bioinformatics [3,26,42], relationship influence in social networks [2,10,11], congestion prediction in traffic network [24], etc. In the literature, uncertain graphs (also known as probabilistic graphs) have been widely utilized to represent these uncertainties [5,47].…”