Hi,
Who writes the integration tests? I can see how useful nightly builds
will be, but you will probably need more than just compiling (if we
have compiled code... ;) and running each set of unit tests.
Cheers,
Michelle
On Tue, 13 Jun 2006, Don Dossa wrote:
Tech Assess concall June 12, 2006
The discussion centered around the expected needs for a software development
system that would be used by pipeline and middleware developers. We discussed
the LSST code development environment.
- Codes are written by individuals at their own institutions on their local
computers. One can easily imagine that this involves around a 100
astronomers/programmers.
- As the codes mature, they will be collected on the LSST development system.
- Since the individuals use whatever platform they chose to write their code,
the first step of the integration is to rebuild their code on the development
system so all codes share a common system environment of operating system,
compilers, libraries, and file system. The developers tests their codes again
and perhaps gain a very crude idea of the code performance on the new platform.
The development system has several ÔloginÕ nodes for this work. Multiple
developers use these login nodes simultaneously.
- On a very regular basis, perhaps nightly, the existing integrated software is
built and tested to the extent possible. Individual developers are responsible
for providing the tests. It may be desirable to have 2 builds - one for the
more stable environment and another for a much more immature integration.
- We discussed how to test those codes which are inherently parallel. Perhaps
we carve out a small ÔdebugÕ partition of multi-processor nodes which
developers can use to test parallel codes. Debug policies are set with some
maximum runtime so everyone gets a chance at the debug partition without an
unreasonable delay. These debug nodes are allocated to one user at a time.
- The majority of the compute nodes test the correct operation of much larger
scale applications and gain a more precise measurement of their performance.
Again, these nodes are dedicated to a single user at a time. Let me call this
the ÔlargeÕ partition.
- Both parallel environments would be used by individual programmers and the
system integration team tying pipelines together with middleware.
- There will be occasions when the system integration team will need nearly the
entire development system to measure the performance of the existing DM
pipelines comprising computing, databases, and middleware.
- We discussed the need to have at least 2 and potentially 3 separate file
systems of increasingly greater size so that preliminary testing does not
impact performance testing on the ÔlargeÕ partition. These individual file
systems would be on separate disks using separate host disk controllers.
- To run the largest possible software functional and performance test, we will
need some mechanism to connect all the file system servers to the large disk
farm.
- Another work item is defining an acceptance test for the hardware at the
development center as well as the summit/base/archive sites. Passing a
stringent hardware acceptance test means we have reliable hardware. Suppliers
are motivated to meet our criteria by including the definition of the
acceptance test in the contract with the provision that they are not paid until
they pass it.
- An equivalent acceptance test should be defined for our integrated software
environment so the project team can track our progress and understand what we
would be ÔdeliveringÕ to them.
- We will continue this quite fascinating and enlightening discussion at the
next tech assess call on June 19.
- Additional discussions outside of the concall concern the floor space, power
and cooling needs of the summit and base computers. Victor K. is expected to
be in La Serena during the week of July 17. I am trying to arrange my schedule
to be there at the same time and also bring along one of LLNLÕs senior facility
managers who is an expert in specify, installing, and debugging large computer
rooms.
Don
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
