Added initial benchmarks for deduplicate filter.

[bwplotka]

The part we did together with @khyatisoneji (pair programming)

The script I used:

#!/usr/bin/env bash

DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"

cd ${DIR}/../

RUN="${1}-dedupfilter"
mkdir -p ${DIR}/bench_outs/${RUN}

for TEST in BenchmarkDeduplicateFilter_Filter/v0 BenchmarkDeduplicateFilter_Filter/v1
do
echo "Running ${TEST} run ${RUN}"
go test -bench=${TEST} -run=^$ -benchmem -memprofile ${DIR}/bench_outs/${RUN}/memprofile${TEST//\//-}.out -timeout 10m -benchtime 30s ./pkg/block | tee  ${DIR}/bench_outs/${RUN}/bench${TEST//\//-}.out
done

sed 's/BenchmarkDeduplicateFilter_Filter\/v0/BenchmarkDeduplicateFilter_Filter/g' ${DIR}/bench_outs/${RUN}/benchBenchmarkDeduplicateFilter_Filter-v0.out > ${DIR}/bench_outs/${RUN}/bench_new.out
sed 's/BenchmarkDeduplicateFilter_Filter\/v1/BenchmarkDeduplicateFilter_Filter/g' ${DIR}/bench_outs/${RUN}/benchBenchmarkDeduplicateFilter_Filter-v1.out > ${DIR}/bench_outs/${RUN}/bench_old.out

benchcmp ${DIR}/bench_outs/${RUN}/bench_old.out ${DIR}/bench_outs/${RUN}/bench_new.out | tee ${DIR}/bench_outs/${RUN}/bench.out

So far @khyatisoneji it looks like it's CPU vs memory tradeoff (: I like that your algorithm has less allocations, but it's quite slow. Not sure yet what wins. I think we can improve your algorithm even more so we can save some CPU. We could do more CPU profiling while benchmarking.

benchmark                                    old ns/op      new ns/op     delta
BenchmarkDeduplicateFilter_Filter/#00-12     571441451      18737137      -96.72%
BenchmarkDeduplicateFilter_Filter/#01-12     5201142904     72444063      -98.61%

benchmark                                    old allocs     new allocs     delta
BenchmarkDeduplicateFilter_Filter/#00-12     3050           4111           +34.79%
BenchmarkDeduplicateFilter_Filter/#01-12     10502          9753           -7.13%

benchmark                                    old bytes     new bytes     delta
BenchmarkDeduplicateFilter_Filter/#00-12     243936        10898912      +4367.94%
BenchmarkDeduplicateFilter_Filter/#01-12     840074        41400505      +4828.20%

TODO(@khyatisoneji ): Make it parallel for resolution (!): For test case number 2.

Pushed that to my fork and will add write access to my fork to you @khyatisoneji, so you can play with it as well!

Signed-off-by: Bartlomiej Plotka [email protected]

[bwplotka]

No longer working on this. We need this, but help wanted for someone to finish it (:

Maybe @khyatisoneji (:

cc #2174