Filebench News

[edit] What's New

FileBench is now included in Open Solaris. It is located in a new /usr/benchmarks directory, which will eventually host some other benchmarks as well. As part of the work to integrate FileBench with Open Solaris, quite a few bugs were fixed, including a number that caused core dumps and one that caused the oltp.f workload to hang during the shutdown phase. The error reporting was also improved, and comment blocks were added to each subroutine which summarize their functionality. The new source code is available at OpenSolaris.org in both downloadable and web browsable form. It is also available for download at SourceForge.org.

Two new pages have been added, one describing the .f workload modeling language in detail, and the other describing the inner workings of FileBench.

FileBench now automatically implements "files" as "filesets" with one entry and unifies the file and fileset access and I/O buffer memory allocation of the read, write, aiowrite, readwholefile, writewholefile, appendfile and appendfilerand flowops. The workload files in filebench/workloads have been updated so that if they are used FileBench will behave as before, but if you want to see how to take advantage of the changes, or if you have your written your own workload files, see unification_of_files_and_filesets for more information.

Filebench now has a random variable feature: FileBench Random Variables. These are variables whose values change with each use according to a specified random distribution function. Several different distributions are supported, and more may be added in the future.

FileBench now allows you to write what are effectively inline subroutines: FileBench Composite Flowops. This feature lets you define new mega flowops as a list of built-in flowops. The Composite Flowops can be used within threads just like any other flowop, including being invoked with iteration counts, and have parameters passed in for setting custom attributes.

Multi-Client capability: FileBench Multi-client Support. A code review is in progress for this feature, which will allow horizontal scaling in a client/server system. Multiple clients will be able to access files and filesets on one (or more) file servers, using the client's NFS or CIFS drivers.