Skip to content

Stale wiki pages

tenchd edited this page Aug 5, 2021 · 6 revisions

Aspen

In all subsequent experiments Aspen was built according to the specs of its developers (compiled with g++-7 -fcilkplus with Cilk Plus runtime libraries) and given access to all 48 cores and full memory on the new lab machine.

Performance evaluation with 1000 update batches, unlimited memory

Input Stream Ingestion Speed (# 10^6 updates / sec) Total Runtime (sec)
kron_13_fourth 1.64828 10.597
kron_15_fourth 1.27607 218.918
kron_16_fourth 1.22838 909.619
kron_17_fourth 1.20785 3700.27

Performance evaluation with 1000 update batches, restricted to 16GB memory

Input Stream Ingestion Speed (# 10^6 updates / sec) Total Runtime (sec)
kron_13_fourth 1.47915 11.8092
kron_15_fourth 1.12393 248.551
kron_16_fourth 1.12532 992.933
kron_17_fourth 1.07073 4174.12

How does Aspen's performance respond to changes in batch size?

On the kron_13_fourth stream:

Batch Size (# updates) Ingestion Speed (# 10^6 updates / sec) Total Runtime (sec)
10^3 1.64084 10.645
10^4 2.68386 6.51
10^5 3.07802 5.676
10^6 3.4112 5.122
10^7 3.42795 5.099
Clone this wiki locally