Added benchmark test for the latest version 3.0.0-rc1.
Introduction of the new parameter optimizeForShortStreams in BeginSessionFor made it impossible (or, at least very much non-trivial) to compare 3.0.0-rc1 with prior versions 3.0.0-alphaX or 2.x directly, using the same BenchmarkDotNet project.
Benchmark results from the latest version 3.0.0-rc1 can be cross-referenced with results for previous versions manually.
Major differences in 3.0.0-rc1 compared to previous versions:
3.0.0-rc1 allows to explicitly optimise for desired mode of operation by using optimizeForShortStreams parameter
3.0.0-alpha24 was optimised for long streams
3.0.0-alpha21 was optimised for short streams
Note that 3.0.0-rc1 and 3.0.0-alpha24 have some other optimisations, e.g. reducing the number of network calls to EventStore server. These optimisations however may not directly affect time spent in loading stream or memory allocated, so these improvements will not affect BenchmarkDotNet results.
Results can be interpreted using following categories:
Short event streams, optimizeForShortStreams=true
In this category 3.0.0-rc1 performance is very close to 3.0.0-alpha24, memory allocations numbers are very close, time spent is slightly better in 3.0.0-rc1 (1.3ms) compared to 3.0.0-alpha24 (1.5ms).
Short event streams, optimizeForShortStreams=false
In this category 3.0.0-rc1 performance is better than 3.0.0-alpha21: e.g. for stream length 20 time spent reduced from 2.0ms to around 1.2ms; memory allocations reduces from 270KB to 210KB.
Long event streams, optimizeForShortStreams=true
In this category 3.0.0-rc1 performance is very close to 3.0.0-alpha21, both in terms of time spent and memory allocations.
Worth noting that same as with 3.0.0-alpha21, when loading long streams we see large number of GC Gen1 and Gen2 collections, so memory pressure is high.
Long event streams, optimizeForShortStreams=false
In this category 3.0.0-rc1 performance is very close to 3.0.0-alpha24: memory allocations are pretty much identical; slightly longer time spent loading stream - 813ms for 3.0.0-rc1 vs 756ms for 3.0.0-alpha24.
Added benchmark test for the latest version
3.0.0-rc1.Introduction of the new parameter
optimizeForShortStreamsinBeginSessionFormade it impossible (or, at least very much non-trivial) to compare3.0.0-rc1with prior versions3.0.0-alphaXor2.xdirectly, using the same BenchmarkDotNet project.Benchmark results from the latest version
3.0.0-rc1can be cross-referenced with results for previous versions manually.Results
3.0.0-rc1Previous versions
Major differences in
3.0.0-rc1compared to previous versions:3.0.0-rc1allows to explicitly optimise for desired mode of operation by usingoptimizeForShortStreamsparameter3.0.0-alpha24was optimised for long streams3.0.0-alpha21was optimised for short streamsResults can be interpreted using following categories:
Short event streams,
optimizeForShortStreams=trueIn this category
3.0.0-rc1performance is very close to3.0.0-alpha24, memory allocations numbers are very close, time spent is slightly better in3.0.0-rc1(1.3ms) compared to3.0.0-alpha24(1.5ms).Short event streams,
optimizeForShortStreams=falseIn this category
3.0.0-rc1performance is better than3.0.0-alpha21: e.g. for stream length 20 time spent reduced from 2.0ms to around 1.2ms; memory allocations reduces from 270KB to 210KB.Long event streams,
optimizeForShortStreams=trueIn this category
3.0.0-rc1performance is very close to3.0.0-alpha21, both in terms of time spent and memory allocations.Worth noting that same as with
3.0.0-alpha21, when loading long streams we see large number of GC Gen1 and Gen2 collections, so memory pressure is high.Long event streams,
optimizeForShortStreams=falseIn this category
3.0.0-rc1performance is very close to3.0.0-alpha24: memory allocations are pretty much identical; slightly longer time spent loading stream - 813ms for3.0.0-rc1vs 756ms for3.0.0-alpha24.Comparing versions prior to the
3.0.0-rc1