[systemd-devel] journal-upload benchmarks

[Zbigniew Jędrzejewski-Szmek]

Hi,

as a rough guide:

src/journal-remote/log-generator.py 10000 > f

NO compression:

time nc --send-only -v localhost 2001 < f
177029931 bytes  2.25s 2.23s 2.216s

time build/systemd-journal-upload -u http://localhost:2001 - < f
177029931 bytes  2.143s 2.100s 2.125s

time build/systemd-journal-upload -u https://localhost:2001 --key certs2/node.key --cert certs2/node.pem --trust certs2/ca.pem - <f
177029931 bytes  3.912s 3.466s 3.444s 3.463s 3.545s

XZ compression (new settings):

time nc --send-only -v localhost 2001 < f
177029931 bytes  30.50s 28.97s 29.21s

LZ4 compression:

177029931 bytes  2.55s 2.22s 2.62s

fake mode, with writing to journal file bypassed, but all the rest left intact:

177008573 bytes  0.38s 0.38s 0.37s

The fact that HTTP overhead is negative suggests something is wrong with buffering.
But it still means that this overhead is negligible.

TLS overhead seems acceptable.

XZ compression is still slow, LZ4 compression is OK.

Writing to the journal is the quite heavy.

Zbyszek