Want to sponsor development of Bonnie++? If so see this link for details.

The benchmark category of my blog has posts related to Bonnie++ and other benchamrk development work. It also includes information on Postal my mail server benchmark.

• Introduction
  This benchmark is named Bonnie++, it is based on the Bonnie benchmark written by Tim Bray. I was originally hoping to work with Tim on developing the next version of Bonnie, but we could not agree on the issue of whether C++ should be used in the program. Tim has graciously given me permission to use the name "bonnie++" for my program which is based around his benchmark.
  Bonnie++ adds the facility to test more than 2G of storage on a 32bit machine, and tests for file creat(), stat(), unlink() operations.
  Also it will output in CSV spread-sheet format to standard output. If you use the "-q" option for quiet mode then the human-readable version will go to stderr so redirecting stdout to a file will get only the csv in the file. The program bon_csv2html takes csv format data on stdin and writes a HTML file on standard output which has a nice display of all the data. The program bon_csv2txt takes csv format data on stdin and writes a formatted plain text version on stdout, this was originally written to work with 80 column braille displays, but can also work well in email.
  
• A note on blocking writes
  I have recently added a -b option to cause a fsync() after every write (and a fsync() of the directory after file create or delete). This is what you probably want to do if testing performance of mail or database servers as they like to sync everything. The default is to allow write-back caching in the OS which is what you want if testing performance for copying files, compiling, etc.
  
• Waiting for semaphores
  There is often a need to test multiple types of IO at the same time. This is most important for testing RAID arrays where you will almost never see full performance with only one process active. Bonnie++ 2.0 will address this issue, but there is also a need for more flexibility than the ability to create multiple files in the same directory and fork processes to access them (which is what version 2.0 will do). There is also need to perform tests such as determining whether access to an NFS server will load the system and slow down access to a local hard drive. Christian Kagerhuber contributed the initial code to do semaphores so that several copies of Bonnie++ can be run in a synchronised fashion. This means you can have 8 copies of Bonnie++ doing per-char reads to test out your 8CPU system!
• Summary of tests
  The first 6 tests are from the original Bonnie: Specifically, these are the types of filesystem activity that have been observed to be bottlenecks in I/O-intensive applications, in particular the text database work done in connection with the New Oxford English Dictionary Project at the University of Waterloo.
  It initially performs a series of tests on a file (or files) of known size. By default, that size is 200 Mb (but that's not enough - see below). For each test, Bonnie reports the number of Kilo-bytes processed per elapsed second, and the % CPU usage (sum of user and system). If a size >1G is specified then we will use a number of files of size 1G or less. This way we can use a 32bit program to test machines with 8G of RAM! NB I have not yet tested more than 2100M of file storage. If you test with larger storage then this please send me the results.
  The next 6 tests involve file create/stat/unlink to simulate some operations that are common bottlenecks on large Squid and INN servers, and machines with tens of thousands of mail files in /var/spool/mail.
  In each case, an attempt is made to keep optimizers from noticing it's all bogus. The idea is to make sure that these are real transfers to/from user space to the physical disk.
• Test Details
  
  • The file IO tests are:
    1. Sequential Output
       1. Per-Character. The file is written using the putc() stdio macro. The loop that does the writing should be small enough to fit into any reasonable I-cache. The CPU overhead here is that required to do the stdio code plus the OS file space allocation.
       2. Block. The file is created using write(2). The CPU overhead should be just the OS file space allocation.
       3. Rewrite. Each BUFSIZ of the file is read with read(2), dirtied, and rewritten with write(2), requiring an lseek(2). Since no space allocation is done, and the I/O is well-localized, this should test the effectiveness of the filesystem cache and the speed of data transfer.
    2. Sequential Input
       1. Per-Character. The file is read using the getc() stdio macro. Once again, the inner loop is small. This should exercise only stdio and sequential input.
       2. Block. The file is read using read(2). This should be a very pure test of sequential input performance.
    3. Random Seeks
       This test runs SeekProcCount processes (default 3) in parallel, doing a total of 8000 lseek()s to locations in the file specified by random() in bsd systems, drand48() on sysV systems. In each case, the block is read with read(2). In 10% of cases, it is dirtied and written back with write(2).
       The idea behind the SeekProcCount processes is to make sure there's always a seek queued up.
       AXIOM: For any unix filesystem, the effective number of lseek(2) calls per second declines asymptotically to near 30, once the effect of caching is defeated.
       One thing to note about this is that the number of disks in a RAID set increases the number of seeks. For read using RAID-1 (mirroring) will double the number of seeks. For write using RAID-0 will multiply the number of writes by the number of disks in the RAID-0 set (provided that enough seek processes exist).
       The size of the file has a strong nonlinear effect on the results of this test. Many Unix systems that have the memory available will make aggressive efforts to cache the whole thing, and report random I/O rates in the thousands per second, which is ridiculous. As an extreme example, an IBM RISC 6000 with 64 Mb of memory reported 3,722 per second on a 50 Mb file. Some have argued that bypassing the cache is artificial since the cache is just doing what it's designed to. True, but in any application that requires rapid random access to file(s) significantly larger than main memory which is running on a system which is doing significant other work, the caches will inevitably max out. There is a hard limit hiding behind the cache which has been observed by the author to be of significant importance in many situations - what we are trying to do here is measure that number.
  • The file creation tests use file names with 7 digits numbers and a random number (from 0 to 12) of random alpha-numeric characters. For the sequential tests the random characters in the file name follow the number. For the random tests the random characters are first.
    The sequential tests involve creating the files in numeric order, then stat()ing them in readdir() order (IE the order they are stored in the directory which is very likely to be the same order as which they were created), and deleting them in the same order.
    For the random tests we create the files in an order that will appear random to the file system (the last 7 characters are in numeric order on the files). Then we stat() random files (NB this will return very good results on file systems with sorted directories because not every file will be stat()ed and the cache will be more effective). After that we delete all the files in random order.
    If a maximum size greater than 0 is specified then when each file is created it will have a random amount of data written to it. Then when the file is stat()ed it's data will be read.
• COPYRIGHT NOTICE
  * Copyright © Tim Bray ([email protected]), 1990.
  * Copyright © Russell Coker ([email protected]) 1999.

  I have updated the program, added support for >2G on 32bit machines, and tests for file creation.
  Licensed under the GPL version 2.0.

• DISCLAIMER
  This program is provided AS IS with no warranty of any kind, and
  The author makes no representation with respect to the adequacy of this program for any particular purpose or with respect to its adequacy to produce any particular result, and
  The authors shall not be liable for loss or damage arising out of the use of this program regardless of how sustained, and In no event shall the author be liable for special, direct, indirect or consequential damage, loss, costs or fees or expenses of any nature or kind.

  NB The results of running this program on live server machines can include extremely bad performance of server processes, and excessive consumption of disk space and/or Inodes which may cause the machine to cease performing it's designated tasks. Also the benchmark results are likely to be bad.

  Do not run this program on live production machines.