Skip navigation

Standard Performance Evaluation Corporation

Facebook logo LinkedIn logo Twitter logo
 
 

Errata

(Updated: July 6, 2021)

Errata for SP2

July 6, 2021: A performance neutral patch which fixes a code path where the vendor has modified the storage2020.yaml file and enabled "skip_init= 1". (This is not permitted for publication but is used by vendors for internal testing because it saves time in nightly regression tests.) The use of skip_init had an unintended side effect on workloads that contained attributes that enabled O_DIRECT on file I/O calls. The patch eliminates the side effect where use of skip_init in combination with the use of O_DIRECT caused the transfer sizes to not match the definitions of the workloads.

Bug fix for customers using the custom workload capabilities of SPECsfs2014_SP2. This does not affect any submission, publication or benchmark usage for submission or publication. This only affects custom workload users.

After applying the patch the user will need to rebuild the binaries and distribute these updated binaries to the nodes involved in the testing.

To build on Unix platforms:

	cd $top
 	make clean
  	make

The new binaries will now be in the 'bin' directory.

To build on Windows platforms:
Use Visual studio to build the executables by double clicking on the msbuild/netmist_dist.sln (solution file) select the desired 32 or 64 bit platform, then click on build.

The new binaries will be in the build\dist\windows\$platform directory.

Errata for SP1 and earlier

The following SfsManager and netmist fixes include:

  1. Fix for adding percent complete console messages in Database workload.
  2. Adding console messages from children after the last heartbeat process exits.
  3. Adding TRACEDEBUG option.
  4. Adding minor cleanup in SfsManager.
  5. Adding NO_OP_VALIDATE option for in house testing.

The following patch cleans up an error path and updates a counter that is normally not used, but could be in custom workloads:


The following patch fixes a minor bug where one of the sections of the generated report is incorrectly a duplicate of another.


When one attempts to run the SFS2014 benchmark on an improperly configured system, the benchmark may not shutdown gracefully and may leave zombie processes behind. The fix below is performance neutral and resolves this problem:


None of these errata fixes affect performance or comparability.