[LU-4191] lustre/tests/lnet-selftest.sh: possible misuse of lst --concurrency

• Status: Open
• Project: Lustre
• Type: Bug
• Priority: Minor
• Reporter: Isaac Huang (Inactive)
• Assignee: WC Triage
• Resolution: Unresolved
• Severity: 3

Description

I looked at a lst test script and saw:
lst add_test --batch b --loop 1000 --concurrency 1 --distribute 2:2 --from c --to s brw read check=full size=4k
lst add_test --batch b --loop 1000 --concurrency 8 --distribute 2:2 --from c --to s brw read check=full size=4k

Unless I've missed something, the two commands are equivalent to a single command:
lst add_test --batch b --loop 1000 --concurrency 9 --distribute 2:2 --from c --to s brw read check=full size=4k

While it's not an error, it doesn't make sense to me to issue two commands while one should suffice.

In lustre/tests/lnet-selftest.sh:
lst_CONCR=${lst_CONCR:-"1 2 4 8"}
if [ "$SLOW" = no ]; then
lst_CONCR="1 8"
fi

I think it'd suffice to:
lst_CONCR=${lst_CONCR:-"8"}
if [ "$SLOW" = no ]; then
lst_CONCR="4"
fi

Running multiple tests at concurrency 8 should suffice to saturate the network, and concurrency 4 would make it run twice faster.

Comments

[Isaac Huang (Inactive)]

Also, the script always uses #clients:#servers or #servers:#clients as the distribution. Liang can you please comment if this is right? What do we get with --distribute 3:2 for example?