A Novel Cache Architecture and Placement Framework for Packet Forwarding Engines

Packet forwarding is a memory-intensive application requiring multiple accesses through a trie structure. With the requirement to process packets at line rates, high-performance routers need to forward millions of packets every second with each packet needing up to seven memory accesses. Earlier work shows that a single cache for the nodes of a trie can reduce the number of external memory accesses. It is observed that the locality characteristics of the level-one nodes of a trie are significantly different from those of lower level nodes. Hence, we propose a heterogeneously segmented cache architecture (HSCA) which uses separate caches for level-one and lower level nodes, each with carefully chosen sizes. Besides reducing misses, segmenting the cache allows us to focus on optimizing the more frequently accessed level-one node segment. We find that due to the nonuniform distribution of nodes among cache sets, the level-one nodes cache is susceptible t high conflict misses. We reduce conflict misses by introducing a novel two-level mapping-based cache placement framework. We also propose an elegant way to fit the modified placement function into the cache organization with minimal increase in access time. Further, we propose an attribute preserving trace generation methodology which emulates real traces and can generate traces with varying locality. Performanc results reveal that our HSCA scheme results in a 32 percent speedup in average memory access time over a unified nodes cache. Also, HSC outperforms IHARC, a cache for lookup results, with as high as a 10-fold speedup in average memory access time. Two-level mappin further enhances the performance of the base HSCA by up to 13 percent leading to an overall improvement of up to 40 percent over the unified scheme.

A Petri net model for evaluating packet buffering strategies in a network processor

Previous studies have shown that buffering packets in DRAM is a performance bottleneck. In order to understand the impediments in accessing the DRAM, we developed a detailed Petri net model of IP forwarding application on IXP2400 that models the different levels of the memory hierarchy. The cell based interface used to receive and transmit packets in a network processor leads to some small size DRAM accesses. Such narrow accesses to the DRAM expose the bank access latency, reducing the bandwidth that can be realized. With real traces up to 30% of the accesses are smaller than the cell size, resulting in 7.7% reduction in DRAM bandwidth. To overcome this problem, we propose buffering these small chunks of data in the on chip scratchpad memory. This scheme also exploits greater degree of parallelism between different levels of the memory hierarchy. Using real traces from the internet, we show that the transmit rate can be improved by an average of 21% over the base scheme without the use of additional hardware. Further, the impact of different traffic patterns on the network processor resources is studied. Under real traffic conditions, we show that the data bus which connects the off-chip packet buffer to the micro-engines, is the obstacle in achieving higher throughput.

Deep Packet Inspection Using Message Passing Networks

We propose a solution based on message passing bipartite networks, for deep packet inspection, which addresses both speed and memory issues, which are limiting factors in current solutions. We report on a preliminary implementation and propose a parallel architecture.

Analytical models for capacity estimation of IEEE 802.11 WLANs using DCF for internet applications

We provide analytical models for capacity evaluation of an infrastructure IEEE 802.11 based network carrying TCP controlled file downloads or full-duplex packet telephone calls. In each case the analytical models utilize the attempt probabilities from a well known fixed-point based saturation analysis. For TCP controlled file downloads, following Bruno et al. (In Networking '04, LNCS 2042, pp. 626-637), we model the number of wireless stations (STAs) with ACKs as a Markov renewal process embedded at packet success instants. In our work, analysis of the evolution between the embedded instants is done by using saturation analysis to provide state dependent attempt probabilities. We show that in spite of its simplicity, our model works well, by comparing various simulated quantities, such as collision probability, with values predicted from our model. Next we consider N constant bit rate VoIP calls terminating at N STAs. We model the number of STAs that have an up-link voice packet as a Markov renewal process embedded at so called channel slot boundaries. Analysis of the evolution over a channel slot is done using saturation analysis as before. We find that again the AP is the bottleneck, and the system can support (in the sense of a bound on the probability of delay exceeding a given value) a number of calls less than that at which the arrival rate into the AP exceeds the average service rate applied to the AP. Finally, we extend the analytical model for VoIP calls to determine the call capacity of an 802.11b WLAN in a situation where VoIP calls originate from two different types of coders. We consider N-1 calls originating from Type 1 codecs and N-2 calls originating from Type 2 codecs. For G711 and G729 voice coders, we show that the analytical model again provides accurate results in comparison with simulations.

Link failure detection for maintaining session continuity in packet data networks

A link failure in the path of a virtual circuit in a packet data network will lead to premature disconnection of the circuit by the end-points. A soft failure will result in degraded throughput over the virtual circuit. If these failures can be detected quickly and reliably, then appropriate rerouteing strategies can automatically reroute the virtual circuits that use the failed facility. In this paper, we develop a methodology for analysing and designing failure detection schemes for digital facilities. Based on errored second data, we develop a Markov model for the error and failure behaviour of a T1 trunk. The performance of a detection scheme is characterized by its false alarm probability and the detection delay. Using the Markov model, we analyse the performance of detection schemes that use physical layer or link layer information. The schemes basically rely upon detecting the occurrence of severely errored seconds (SESs). A failure is declared when a counter, that is driven by the occurrence of SESs, reaches a certain threshold.For hard failures, the design problem reduces to a proper choice;of the threshold at which failure is declared, and on the connection reattempt parameters of the virtual circuit end-point session recovery procedures. For soft failures, the performance of a detection scheme depends, in addition, on how long and how frequent the error bursts are in a given failure mode. We also propose and analyse a novel Level 2 detection scheme that relies only upon anomalies observable at Level 2, i.e. CRC failures and idle-fill flag errors. Our results suggest that Level 2 schemes that perform as well as Level 1 schemes are possible.

Packet Reordering in Network Processors

Network processors today consist of multiple parallel processors (micro engines) with support for multiple threads to exploit packet level parallelism inherent in network workloads. With such concurrency, packet ordering at the output of the network processor cannot be guaranteed. This paper studies the effect of concurrency in network processors on packet ordering. We use a validated Petri net model of a commercial network processor, Intel IXP 2400, to determine the extent of packet reordering for IPv4 forwarding application. Our study indicates that in addition to the parallel processing in the network processor, the allocation scheme for the transmit buffer also adversely impacts packet ordering. In particular, our results reveal that these packet reordering results in a packet retransmission rate of up to 61%. We explore different transmit buffer allocation schemes namely, contiguous, strided, local, and global which reduces the packet retransmission to 24%. We propose an alternative scheme, packet sort, which guarantees complete packet ordering while achieving a throughput of 2.5 Gbps. Further, packet sort outperforms the in-built packet ordering schemes in the IXP processor by up to 35%.

Two-level Mapping Based Cache Index Selection for Packet Forwarding Engines

Packet forwarding is a memory-intensive application requiring multiple accesses through a trie structure. The efficiency of a cache for this application critically depends on the placement function to reduce conflict misses. Traditional placement functions use a one-level mapping that naively partitions trie-nodes into cache sets. However, as a significant percentage of trie nodes are not useful, these schemes suffer from a non-uniform distribution of useful nodes to sets. This in turn results in increased conflict misses. Newer organizations such as variable associativity caches achieve flexibility in placement at the expense of increased hit-latency. This makes them unsuitable for L1 caches.We propose a novel two-level mapping framework that retains the hit-latency of one-level mapping yet incurs fewer conflict misses. This is achieved by introducing a secondlevel mapping which reorganizes the nodes in the naive initial partitions into refined partitions with near-uniform distribution of nodes. Further as this remapping is accomplished by simply adapting the index bits to a given routing table the hit-latency is not affected. We propose three new schemes which result in up to 16% reduction in the number of misses and 13% speedup in memory access time. In comparison, an XOR-based placement scheme known to perform extremely well for general purpose architectures, can obtain up to 2% speedup in memory access time.

State Dependent Attempt Rate modeling of single cell IEEE 802.11 WLANs with homogeneous nodes and Poisson packet arrivals

We study a State Dependent Attempt Rate (SDAR) approximation to model M queues (one queue per node) served by the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol as standardized in the IEEE 802.11 Distributed Coordination Function (DCF). The approximation is that, when n of the M queues are non-empty, the (transmission) attempt probability of each of the n non-empty nodes is given by the long-term (transmission) attempt probability of n saturated nodes. With the arrival of packets into the M queues according to independent Poisson processes, the SDAR approximation reduces a single cell with non-saturated nodes to a Markovian coupled queueing system. We provide a sufficient condition under which the joint queue length Markov chain is positive recurrent. For the symmetric case of equal arrival rates and finite and equal buffers, we develop an iterative method which leads to accurate predictions for important performance measures such as collision probability, throughput and mean packet delay. We replace the MAC layer with the SDAR model of contention by modifying the NS-2 source code pertaining to the MAC layer, keeping all other layers unchanged. By this model-based simulation technique at the MAC layer, we achieve speed-ups (w.r.t. MAC layer operations) up to 5.4. Through extensive model-based simulations and numerical results, we show that the SDAR model is an accurate model for the DCF MAC protocol in single cells. (C) 2012 Elsevier B.V. All rights reserved.

Packet scheduling for priority based transmission in energy harvesting sensors

In this paper, we determine packet scheduling policies for efficient power management in Energy Harvesting Sensors (EHS) which have to transmit packets of high and low priorities over a fading channel. We assume that incoming packets are stored in a buffer and the quality of service for a particular type of message is determined by the expected waiting time of packets of that type of message. The sensors are constrained to work with the energy that they garner from the environment. We derive transmit policies which minimize the sum of expected waiting times of the two types of messages, weighted by penalties. First, we show that for schemes with a constant rate of transmission, under a decoupling approximation, a form of truncated channel inversion is optimal. Using this result, we derive optimal solutions that minimize the weighted sum of the waiting times in the different queues.

Network properties of decoys and CASP predicted models: a comparison with native protein structures

Protein structure space is believed to consist of a finite set of discrete folds, unlike the protein sequence space which is astronomically large, indicating that proteins from the available sequence space are likely to adopt one of the many folds already observed. In spite of extensive sequence-structure correlation data, protein structure prediction still remains an open question with researchers having tried different approaches (experimental as well as computational). One of the challenges of protein structure prediction is to identify the native protein structures from a milieu of decoys/models. In this work, a rigorous investigation of Protein Structure Networks (PSNs) has been performed to detect native structures from decoys/ models. Ninety four parameters obtained from network studies have been optimally combined with Support Vector Machines (SVM) to derive a general metric to distinguish decoys/models from the native protein structures with an accuracy of 94.11%. Recently, for the first time in the literature we had shown that PSN has the capability to distinguish native proteins from decoys. A major difference between the present work and the previous study is to explore the transition profiles at different strengths of non-covalent interactions and SVM has indeed identified this as an important parameter. Additionally, the SVM trained algorithm is also applied to the recent CASP10 predicted models. The novelty of the network approach is that it is based on general network properties of native protein structures and that a given model can be assessed independent of any reference structure. Thus, the approach presented in this paper can be valuable in validating the predicted structures. A web-server has been developed for this purpose and is freely available at http://vishgraph.mbu.iisc.ernet.in/GraProStr/PSN-QA.html.

Packet flow analysis in IP networks using data-flow analysis

Static analysis (aka offline analysis) of a model of an IP network is useful for understanding, debugging, and verifying packet flow properties of the network. Data-flow analysis is a method that has typically been applied to static analysis of programs. We propose a new, data-flow based approach for static analysis of packet flows in networks. We also investigate an application of our analysis to the problem of inferring a high-level policy from the network, which has been addressed in the past only for a single router.

Synconset Waves and Chains: Spiking Onsets in Synchronous Populations Predict and Are Predicted by Network Structure

Synfire waves are propagating spike packets in synfire chains, which are feedforward chains embedded in random networks. Although synfire waves have proved to be effective quantification for network activity with clear relations to network structure, their utilities are largely limited to feedforward networks with low background activity. To overcome these shortcomings, we describe a novel generalisation of synfire waves, and define `synconset wave' as a cascade of first spikes within a synchronisation event. Synconset waves would occur in `synconset chains', which are feedforward chains embedded in possibly heavily recurrent networks with heavy background activity. We probed the utility of synconset waves using simulation of single compartment neuron network models with biophysically realistic conductances, and demonstrated that the spread of synconset waves directly follows from the network connectivity matrix and is modulated by top-down inputs and the resultant oscillations. Such synconset profiles lend intuitive insights into network organisation in terms of connection probabilities between various network regions rather than an adjacency matrix. To test this intuition, we develop a Bayesian likelihood function that quantifies the probability that an observed synfire wave was caused by a given network. Further, we demonstrate it's utility in the inverse problem of identifying the network that caused a given synfire wave. This method was effective even in highly subsampled networks where only a small subset of neurons were accessible, thus showing it's utility in experimental estimation of connectomes in real neuronal-networks. Together, we propose synconset chains/waves as an effective framework for understanding the impact of network structure on function, and as a step towards developing physiology-driven network identification methods. Finally, as synconset chains extend the utilities of synfire chains to arbitrary networks, we suggest utilities of our framework to several aspects of network physiology including cell assemblies, population codes, and oscillatory synchrony.