This is unexpected if one wants to test performance of a standard filesystem (by pointing replay_redirect to a standard file) with buffered io. Signed-off-by: Lukas Straub <lukasstraub2@xxxxxx> --- blktrace.c | 8 -------- 1 file changed, 8 deletions(-) diff --git a/blktrace.c b/blktrace.c index f1dbd1a6..7682a4d5 100644 --- a/blktrace.c +++ b/blktrace.c @@ -582,14 +582,6 @@ bool read_blktrace(struct thread_data* td) td->o.max_bs[DDIR_TRIM] = rw_bs[DDIR_TRIM]; } - /* - * We need to do direct/raw ios to the device, to avoid getting - * read-ahead in our way. But only do so if the minimum block size - * is a multiple of 4k, otherwise we don't know if it's safe to do so. - */ - if (!fio_option_is_set(&td->o, odirect) && !(td_min_bs(td) & 4095)) - td->o.odirect = 1; - /* * If depth wasn't manually set, use probed depth */ -- 2.34.1
Attachment:
pgprhg_rL50zI.pgp
Description: OpenPGP digital signature
