Automatic specialization of protocol stacks in operating system kernels

Abstract

Abstract-Fast and optimized protocol stacks play a major role in the performance of network services. This role is especially important in embedded class systems, where performance metrics such as data throughput tend to be limited by the CPU. It is common on such systems, to have protocol stacks that are optimized by hand for better performance and smaller code footprint. In this paper, we propose a strategy to automate this process.Our approach uses program specialization, and enables applications using the network to request specialized code based on the current usage scenario. The specialized code is generated dynamically and loaded in the kernel to be used by the application.We have successfully applied our approach to the TCP/IP implementation in the Linux kernel and used the optimized protocol stack in existing applications. These applications were minimally modified to request the specialization of code based on the current usage context, and to use the specialized code generated instead of its generic version. Specialization can be performed locally, or deferred to a remote specialization server using a novel mechanism [1] Experiments conducted on three platforms show that the specialized code runs about 25% faster and its size reduces by up to 20 times. The throughput of the protocol stack improves by up to 21%.

vSpec: workload-adaptive operating system specialization for virtual machines in cloud computing

“…The authors of [13] collected running metrics and classified workloads into web server, FTP server and database server with a support vector machine (SVM). Jiang et al [14] classified applications through flow-level statistics based directly upon standard NetFlow 1) records, which used a naive Bayes classification method. Zhang et al [15] used principal component analysis to decrease the dimension of the metrics collected and adopted the K-nearest neighbors algorithm to classify workloads.…”

“…Pu et al [35] specialized the Unix file system incrementally and improved kernel call performance based on re-plugging. The authors of [36,37] optimized the protocol stack for a TCP/IP implementation. Marinos et al [38] optimized the stack for static web content and DNS server, resulting in higher web-server and DNS throughput and lower CPU usage.…”

“…The common OS indiscriminately allocates the capacity of the CPU, memory, I/O and network. In many applications, the dominant kernel functions and system configuration are different [1,2], so an OS designed as suit-to-all works well in some situations while it is ill-suited in others [3][4][5][6]. The conflict between generality and performance has become a stumbling block to high-performance cloud computing [7].…”

Isolation points

“…Such methods, therefore, are usefully employed with the isolation point-based approach advocated in this paper. Also of use is earlier work on dynamic system adaptation [Bhatia et al 2004;Cowan et al 1995;Rosu et al 1998;Yuan and Nahrstedt 2004], which attempts to improve the behavior of the distributed systems' performance [Noble et al 1997;Rosu et al 1998]. …”

Shadow Kernels

“…Request permissions from permissions@acm.org. At the same time, kernel specialization has been shown to be beneficial [6,19]: profile-guided optimization of Linux can improve performance by up to 10% for some applications [20], exokernels eliminate kernel abstractions for applications so that applications communicate more directly with hardware thereby reducing kernel overheads [7], and kernel instrumentation can be added that only fires when the kernel is executing on behalf of certain processes.…”