Issues with Selected Scalability Features of the 2.6 Kernel

The 2.6 kernel has a number of features that improve performance on high-end SMP and NUMA systems. Finer-grain locking is used in the scheduler, the block I/O layer, hardware and software interrupts, memory management, and the VFS layer. In addition, 2.6 brings new primitives such as RCU and per-cpu data, lock-free algorithms for route cache and directory entry cache as well as scalable user-level APIs like sys_epoll() and futexes. With the widespread testing of these features of the 2.6 kernel, a number of new issues have come to light that needs careful analysis. Some of these issues encountered thus far are - overhead of multilple lock acquisitions and atomic operations in critical paths, possibility of denial-of-service attack on subsystems that use RCU-based deferred free algorithms and degradation of realtime response due to increased softirq load.

In this paper, we analyze a select set of these issues, present the results, workaround patches where exist and future courses of action. We also discuss applicability of some these issues in new features being planned for 2.7 kernel.