Florian A. Auernhammer

The cache and memory subsystems of the IBM POWER8 processor

In this paper, we describe the IBM POWER8™ cache, interconnect, memory, and input/output subsystems, collectively referred to as the “nest.” This paper focuses on the enhancements made to the nest to achieve balanced and scalable designs, ranging from small 12-core single-socket systems, up to large 16-processor-socket, 192-core enterprise rack servers. A key aspect of the design has been increasing the end-to-end data and coherence bandwidth of the system, now featuring more than twice the bandwidth of the POWER7® processor. The paper describes the new memory-buffer chip, called Centaur, providing up to 128 MB of eDRAM (embedded dynamic random-access memory) buffer cache per processor, along with an improved DRAM (dynamic random-access memory) scheduler with support for prefetch and write optimizations, providing industry-leading memory bandwidth combined with low memory latency. It also describes new coherence-transport enhancements and the transition to directly integrated PCIe® (PCI Express®) support, as well as additions to the cache subsystem to support higher levels of virtualization and scalability including snoop filtering and cache sharing.

Design optimization of a highly parallel InfiniBand host channel adapter

Network processors use highly parallel architectures to improve performance and reach multi-gigabit line-speeds. In this paper, we emulate a pipeline in a highly parallel non-programmable industrial InfiniBand Host Channel Adapter to make a performance and bottleneck analysis and, at the same time, explore the potential of a pipelined architecture. Therefore, starting from the original Host Channel Adapter model with multiple send- and receive-side packet-processing units, we compare its performance capabilities with that of a pipelined design by introducing a central arbiter synchronizing the state machines of the different packet-processing instances to achieve a pipelined behavior. We show that the pipelined model achieves a performance comparable to that of the parallel design in most of our micro-benchmarks, making it a valid option for next-generation high-speed adapters. At the same time, our approach enables a deeper analysis of the original architecture and a better understanding of the actual processing requirements, and therefore offers valuable insights for future designs.