File System Multithreading in System V Release 4 MP

An Intel-sponsored Consortium of computer companies has developed a multiprocessor version of System V Release 4 (SVR4MP) which has been released by UNIX System Laboratories. The Consortium's goal was to add fine-grained locking to SVR4 with minimal change to the kemel, and with complete backward compatibility for user programs. To do this, a locking strategy was developed which complemented, rather than replaced, existing UNIX synchronization mechanisms.

To multithread the flle systems, some general locking strategies were developed and applied to the generic Virtual File System (VFS) and vnode interfaces. Of particular interest were the disk-based S5 and UFS file system types, especially with respect to their scalability. Contention points were found and successively eliminated to the point where the the systems were found to be disk-bound. In particular, several file system caches were restructured using a low-contention, highly-scalable approach called a Software Set-Associative Cache. This technique reduced the measured locking contention of each of these caches from the 10-15% range to less than 0.1%.

A number of experimental changes to disk queue sorting algorithms were attempted to reduce the disk bottleneck, with limited success. However, these experiments provided the following insight into the need for balance between I/O and CPU utilization in the system: that attempting to increase CPU utilization to show higher parallelism could actually lower system throughput.

Using the GAEDE benchmark with a sufflcient number of disks configured, the kernel was found to obtain throughput scalability of 88% of the theoretical maximum on 5 processors.