[SYSTEMDS-3862] SSB Benchmark Implementation #2280
Conversation
ghafek
force-pushed
the
feature/ssb-benchmark
branch
from
bcc4671 to
e7ae6b3
Compare
ghafek
changed the title
There was a problem hiding this comment.
Please, move this from shell/run_all_perf.sh to scripts/ssb/shell/run_all_perf.sh
There was a problem hiding this comment.
Please, move this from shell/run_ssb.sh to scripts/ssb/shell/run_ssb.sh
There was a problem hiding this comment.
Please, move this and all the other .sql files from sql/ to scripts/ssb/sql/
run_all_perf.sh
Outdated
There was a problem hiding this comment.
Please, remove this file from repository root
run_ssb.sh
Outdated
There was a problem hiding this comment.
Please, remove this file from repository root
This patch refines the current union operation to an internal LOP operation. Currently, two subsequent operations -- rbind() and unique() are used to perform the union operation. We rewrite the operation with an internal LOP that uses a HashSet to compute the unique entries and returns them in a matrix. This improves the efficiency of the operation, as it avoids unique(). The order of the input entries is preserved in the output. Closes apache#2286.
This patch introduces a basic integration of the out-of-core backend. For reading, we use a dedicated reblock instruction which creates a queue of blocks, spawns a thread for reading and immediately returns. In addition, we extended the acquireRead functionality to collect such streams of blocks whenever an operations requires the full matrix. Based on these foundations, we can now add other OCC operations that directly work with the input stream of blocks and produce either results or created modified output streams.
The test failure of not finding the mtd-file did not show up in local tests, because locally we do not clean the test directories, and from development there were both Xmtd and X.mtd existing.
This patch introduces the out-of-core unary aggregate operations as an example of how to implement operations against the input stream of blocks.
- Added SSB (Star Schema Benchmark) query implementations - Created performance testing framework with run_all_perf.sh - Added data caching and preprocessing capabilities - Implemented comprehensive logging and output management - Added documentation and status tracking files This is work in progress for the SSB benchmark feature.
ghafek
force-pushed
the
feature/ssb-benchmark
branch
from
2ebb5d1 to
fd7edeb
Compare
|
Hi @gaturchenko , I hope you're doing well. I wanted to let you know that all merge conflicts have now been resolved and the feature branch is clean again. The requested changes from your previous review have been implemented as well. Before proceeding further, I would like to ask for your preference: Would you prefer 1.to continue the review on this existing PR, 2.that I create a new, clean PR based on an up-to-date main branch (containing only the SSB-related commits and no historical noise)? I'm completely fine with both options — just let me know what works best for you. Thank you again for your time, review, and guidance! |
6 participants
No description provided.