We consider cluster-based network servers in which a front-end directs incoming requests to one of a number of back-ends. Specifically, we consider content-based request distribution:

We consider cluster-based network servers in which a front-end directs incoming requests to one of a number of back-ends. Speci cally, w e consider content-based request distribution: the front-end uses the content r equested, in addition to information about the load on the back-end nodes, to choose which b a c k-end will handle this request. Content-based request distribution can improve locality in the back-ends' main memory caches, increase secondary storage scalability b y partitioning the server's database, and provide the ability to employ back-end nodes that are specialized for certain types of requests.As a speci c policy for content-based request distribution, we i n troduce a simple, practical strategy for locality-aware request distribution (LARD). With LARD, the front-end distributes incoming requests in a manner that achieves high locality in the back-ends' main memory caches as well as load balancing. Locality is increased by dynamically subdividing the server's working set over the back-ends. Trace-based simulation results and measurements on a prototype implementation demonstrate substantial performance improvements over state-of-the-art approaches that use only load information to distribute requests. On workloads with working sets that do not t in a single server node's main memory cache, the achieved throughput exceeds that of the state-of-the-art approach b y a factor of two to four.With content-based distribution, incoming requests must be handed o to a back-end in a manner transparent to the client, after the front-end has inspected the content of the request. To this end, we i n troduce an e cient TCP hando protocol that can hand o an established TCP connection in a client-transparent manner.To appear in the Proceedings of the Eighth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS-VIII), San Jose, CA, Oct 1998. IntroductionNetwork servers based on clusters of commodity w orkstations or PCs connected by high-speed LANs combine cutting-edge performance and low cost. A cluster-based network server consists of a front-end, responsible for request distribution, and a number of back-end nodes, responsible for request processing. The use of a front-end makes the distributed nature of the server transparent to the clients. In most current cluster servers the frontend distributes requests to back-end nodes without regard to the type of service or the content requested. That is, all back-end nodes are considered equally capable of serving a given request and the only factor guiding the request distribution is the current load of the backend nodes.With content-based r equest distribution, the frontend takes into account both the service/content r equested and the current load on the back-end nodes when deciding which back-end node should serve a given request. The potential advantages of content-based request distribution are: (1) increased performance due to improved hit rates in the back-end's main memory caches, (2) increased secon...

Given the complexity and associated cost of building modern computer systems, simulation is often the only practical way to test architectural ideas and assess system performance. Simulators provide the flexibility to modify and analyze the impact of various architectural parameters and components as well as enable more detailed statistics collection than real hardware. These benefits make simulation useful even for projects that will eventually implement hardware.Prior to 1994, most academic shared-memory multiprocessor studies largely ignored the processor model, focusing instead on the memory system as the most important performance bottleneck. These studies assumed a simplistic processor model based on in-order issue, blocking reads, and no speculation. However, the early 1990s saw several announcements of commercial shared-memory systems using processors that aggressively exploited instruction-level parallelism (ILP) such as the MIPS R10000, Hewlett-Packard PA8000, and Intel Pentium Pro. These processors had the potential to reduce memory read stalls by overlapping read latency with other operations, possibly changing the nature of performance bottlenecks in the system.Because no shared-memory ILP systems or simulators were available at that time, we designed Rsim-originally an acronym for Rice simulator for ILP multiprocessors-to study such systems. Two major questions guided our efforts:• Does processor microarchitecture influence shared-memory performance and design to the extent that it justifies its detailed modeling and associated performance costs in a shared-memory simulator? • With simple processor-based simulators already taking a long time to run, could we build such a detailed simulator efficiently enough to perform substantive architecture studies in reasonable time?Our experience with Rsim demonstrates that modeling ILP features is important even in sharedmemory multiprocessor systems. In particular, current simple processor-based approximations cannot model significant performance effects for applications exhibiting parallel read misses. Further, recent shared-memory designs-for example, aggressive implementations of sequential consistency 1 -directly use the aggressive ILP-enhancing features of modern processors that simple processor-based simulators do not model.We have also demonstrated that significant multiprocessor studies can be performed with the current speed of ILP simulators. However, improving their speed is crucial for future workloads. Our Rsim is a publicly available architecture simulator for shared-memory systems built from processors that aggressively exploit instruction-level parallelism. Modeling ILP features in a multiprocessor is particularly important for applications that exhibit parallelism among read misses.

No abstract

No abstract