Aurora Release Notes

Aurora 1.0.6

Branches Merged In

• branch_operators_vldbjournal
• (various fixes / improvements directly on the trunk)
  
  

Enhancements

• Implemented the operators in VLDB Journal's operators paper
  
  
• Added interface for the delayed-statistics code that the Load Shedder will make use of. The interface isn't yet implemented. The interface starts at a class named DelayedDataMgr.
  
  
• Added a new class into the Aurora runtime: RuntimeGlobals.  Code in the Aurora runtime cane use this class as an easy way to access the runtime's TraceLogger, PropsFile, and DelayedDataMgr. We haven't reworked existing code to use this class when appropriate.
  

Older versions' release notes

• 1.0.3
• 1.0.4
• 1.0.5
  

Building the Code

1. Set the AURORA_SRC_DIR environment variable to give the path of the top directory of the aurora source code. I.e., /home/jdoe/aurora/.
   
   
2. Set the AURORA_BUILD_ENV environment variable to one of the valid values spelled out in the top few lines of the file aurora/makedefs_locations.mk.
   
   
3. type cd $AURORA_SRC_DIR
   
   
4. type make
   

Using Aurora

• To run the Aurora development GUI, do this:
1. type cd aurora/gui
2. type ant run
   
   
• To use the Aurora C++ API:
• The header files are found in aurora/auroralib/include
• Link your program to the archive file aurora/auroralib/lib/libaurora.
  

Required 3rd Party Software

Older Versions' Release Notes

Aurora 1.0.3

Branches Merged In

• turn_aurora_into_lib

Change Summary

1. Introduced the Aurora C++ API. This is the new way that applications should interact with the Aurora runtime system. This is in the aurora/auroralib directory tree.
   
   
2. Altered the runtime system to encourage programmers to use a single, runtime-wide properties file for startup parameters.
   
   
3. Introduced a logging subsystem to the runtime. It can be used for both debugging output and normal runtime messages.
   
   
4. Created the top-level directory aurora/util. Contains many utility classes/functions for XML parsing, properties files, threading, etc.
   
   
5. Major re-organization of the Scheduler. Most of the code that used to exist in scheduler/src/main.C has been moved into the initialization method of the Aurora C++ API.
   
   
6. Little code changes throughout the code base to eliminate most compiler warnings.
   
   
7. Removed support for GCC 2.95. The new code makes use of some STL features that only appear in the 3.2.x series of GCC compilers.
   
   
8. The Scheduler used to contain code for automatically generating query network catalogs. This code has been moved into the Catalog Manager subsystem and made its own program. After Aurora has been built, you'll find the program at: aurora/catalogmgr/bin/netgen.
   

Known Bugs / Limitations

1. Because so many people were waiting on this code, I violated my own check-in mandates. While the code compiles cleanly, I didn't subject it to a recent Valgrind analysis. This analysis should be done soon.
   
   
2. The makefile aurora/makedefs_locations.mk still requires custom modification for each different site that wants to build Aurora. This should be replaced by autoconf in the near future to give a more flexible build environment.
   
   
3. The Aurora API archive, when built in debug mode: around 70 MB. We should look into how to reduce this. (Optimized builds get it down to a very acceptable size.)
   
   
4. One of the unit test files requires modification before runtime use, because it contains a site-specific path. The file is aurora/auroralib/src/testloadgen/dataset1_props.xml.
   
   
5. The QoSMonitor and the QueueMonitor functionality, as seen in the original MITRE demos, is broken. This was actually broken prior to the creation of the turn_aurora_into_lib branch. Alex Rasin has expressed some willingness to fix this in the future.
   
   
6. The MITRE demo cannot be built from this code. The MITRE demo source code wasn't being maintained even before the creatio of the turn_aurora_into_lib branch. Now that we have the Aurora API on which to build demos, the fixing of the MITRE demo should probably involve porting it to the demo to the Aurora API.
   
   
7. Filter boxes contain two output ports. The top port produces tuples that match the filter predicate. The bottom port is supposed to produce tuple not matching the filter predicate. However, it simply doesn't produce any tuples at all.
   

Aurora 1.0.4

Branches Merged In

• branch_aurora_c_api1
  
  

Enhancements

• Added a C API for Aurora applications. This is perfectly useable by an application, but the purpose for making it was to lay the groundwork for an Aurora Java API.
  
  
• Added a utility program: aurora/util/bin/alselect. This program takes, via stdin,  a log file generated by Aurora's logging system. This program outputs, on stdout, only the subset of those messages that are in desired message groups.
  
  

Bug Fixes

• Removed some printing by the scheduler to stdout / stderr. Such printing violates our design goals for the library.
  
  

Known Bugs / Limitations

Aurora 1.0.5

Branches Merged In

• branch_faster_builds1
  
  

Enhancements

• Made the runtime's Makefiles default to optimizing the object files, rather than including debug info. This is because:
• Most people don't seem to be using interactive debuggers when working on Aurora C++ code
• Aurora builds ~ 15% faster this way
• Object files' / archives' sizes shrink by about 95%
  
  
• In the runtime's Makefiles, enabled the "-Wuninitialized" flag by default, which means C++ programmers quickly discover some cases of using uninitialized variables.
  
  
• Made a slight improvement in how we build libaurora_c.a and libaurora_cxx.a, slightly reducing the time it takes to produce those files.
  
  
• If you use the programs 'testloadgen' or 'test_capi1' to exercise your code, here's how to do a minimal rebuild when you modify your code:

1. Modify your code. I.e., aurora/sched/src/Scheduler.C
   
   
2. cd .../aurora/auroralib/src/
   
   
3. make

Bug Fixes

• None.
  
  

Known Bugs / Limitations