Markus Dreseler
Markus Dreseler
As @mrzzzrm rightly said in #647, we need a solid way of verifying potential performance improvements. Similar to how the CI tells us how a PR changed the coverage, it...
## Make IndexScan a TableScanImpl If the optimizer decided to use an index, the TableScan should create two impls, where the first is the one that is currently generated and...
If we are on a NUMA system and we don't set anything, the Linux kernel wildly migrates our process. No idea why, and it doesn't really make sense. We should...
In #1980, we found that the JoinHash didn't particularly like the input table growing while the join was executed. Both the chunk count increasing and the size of the last...
Some options should be shared among the binaries for the console, the server, and benchmarks. Maybe even tests. * `--default-compression` * `--scheduler` * `--logger` * ...? Maybe even things like...
The header filter for clang-tidy is not properly set, meaning that errors in hpp files are ignored. Some checks show actual issues: ``` /Users/markus/Projekte/Opossum/Git/build-tidy/../src/lib/utils/singleton.hpp:20:11: error: use '= default' to define...
When limiting the number of cores with `--cores` but not setting `numactl -m`, the parallel encoding might mean that data is spread over all NUMA nodes. This can be good...
Currently, we only have the wall time of the operators. If one operator runs 10 JobTasks, it consumes 10x more resources than the PQP tells us. If possible, we should...
The [compiler support matrix](https://github.com/boostorg/hana/wiki/General-notes-on-compiler-support) says that icc is not supported because it lacks *C++14 variable templates*. Since September, [those are said to be supported by Intel](https://software.intel.com/en-us/articles/c14-features-supported-by-intel-c-compiler). How does that change...
As seen in #111, the print* methods are entirely untested. Since we want this to be useful for people outside of Hyrise, too, we should make sure that we add...