mongoperf

mongoperf is a utility for checking disk i/o performance of a server independent of MongoDB. It performs simple timed random disk i/o's.

The utility is new and will likely be more sophisticated in the future.

# get help: 
mongoperf -h

# example invocation:
echo "{nThreads:16,fileSizeMB:1000,r:true}" | ./mongoperf | tee out

Building

Run scons mongoperf to build.

mmf:false mode (default mode)

In the default mode of operation, random 4KB direct disk i/o's are performed (i.e., O_DIRECT is used on Linux). Thus this is a physical disk i/o test.

Example run

mmf:true mode

If mmf:true is specified as an option, tests are performed using memory-mapped files. These files are openedin a "normal" fashion and thus caching is allowed. This sometimes can be used to test file system cache behavior with memory mapped files.

syncDelay option

The syncDelay:secs option instructs mongoperf to perform an asynchronous fsync of the test mmap file at the specified interval. mongod does something similar every 60 seconds, thus this can be useful to test basic system behavior in a simpler setting. This option is applicable only when using mmf:true mode.

fileSizeMB option

This specifies the size of the test data file; you will need this much free disk space. The file will be placed in the current directory.

Specify a very large test file size to create a realistic simulation. A very small file (1MB) could be cached by your disk controller compltely. A file that is say, 100MB in size would involve a small number of disk cylinders – track-to-track seeks are much faster than a drive's average seek time.

The file size is particularly important when mmf:true is specified, as the file system cache is then involved. A file size much larger than RAM will result in much different performance results than a file size smaller than RAM.