Performance modeling and architecture exploration of network processors

Abstract

This paper proposes a Petri net model for a commercial network processor (Intel IXP architecture) which is a multithreaded multiprocessor architecture. We consider and model three different applications viz., IPv4 forwarding, Network Address Translation, and IP Security running on IXP 2400/2850. A salient feature of the Petri net model is its ability to model the application, architecture and their interaction in great detail. The model is validated using the Intel proprietary tool (SDK 3.51 for IXP architecture) over a range of configurations.We conduct a detailed performance evaluation, identify the bottleneck resource, and propose a few architectural extensions and evaluate them in detail.

A Petri Net Model for Evaluating Packet Buffering Strategies in a Network Processor

“…This allows us to model the bank conflicts during DRAM access. Unlike the model in [9], the Petri net model handles packets of different lengths. We model the cell based interface for receiving and transmitting packets (described in Section 2.1.1), the SRAM and scratchpad memories and the buses connecting them to the MEs.…”

“…Accessing it, results in a narrow access of 24 bytes 1 . Govind, et al [9] propose storing the header in the SRAM in an effort to reduce the fully saturated DRAM utilization. We observe that the packet header is accessed multiple times and exhibits certain amount of locality.…”

“…Exploiting DRAM bandwidth is the key to obtaining higher throughput in network applications [10]. In [9], the authors show that buffering packets in the DRAM is a bottleneck in network processing applications. In order to understand these overheads, we study the memory hierarchy in detail.…”

“…In order to understand these overheads, we study the memory hierarchy in detail. Unlike the probabilistic model used in [9], our colored Petri net model captures the various hardware structures of the DRAM, including banks, rows, columns, page buffers and the data bus. We also model the packet buffer allocation, DRAM address mapping scheme and cell based interface for packet transmission.…”

“…Using the detailed Petri net model for IP forwarding, the throughput of the NP is studied using real traces from the internet. Previous studies [9] only considered traffic consisting of 64 byte sized packets and showed a throughput of 3.2 Gbps. For this trace, limited computation power of the hash unit is observed to be an impediment.…”