From 0c3a5c6dceef19ec8ad6538bd6e414376b9ee922 Mon Sep 17 00:00:00 2001
From: Nitin Garg <113666283+gargnitingoogle@users.noreply.github.com>
Date: Wed, 20 Dec 2023 16:59:17 +0530
Subject: [PATCH] Lower freq of syncs in periodic-perf-tests (#1549)

This is for internal bug id 316284545 .
Currently in the periodic perf tests, we
fsync after each write (set through fsync=1)
in all write tests. This poses two problems.

1. It underrates our write bandwidth as most
of the runtime is spent in the sync calls and
little in the actual write operation.
2. This causes frequent sync which result in
frequent object generations on GCS, which is
throttled by GCS.

This change should address both these problems.
---
 perfmetrics/scripts/job_files/seq_rand_read_write.fio | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/perfmetrics/scripts/job_files/seq_rand_read_write.fio b/perfmetrics/scripts/job_files/seq_rand_read_write.fio
index 894c811586..4810d747a1 100644
--- a/perfmetrics/scripts/job_files/seq_rand_read_write.fio
+++ b/perfmetrics/scripts/job_files/seq_rand_read_write.fio
@@ -34,6 +34,7 @@ directory=gcs/256kb
 filesize=256k
 rw=write
 numjobs=40
+fsync=16
 
 [3_thread]
 stonewall
@@ -49,6 +50,7 @@ directory=gcs/3mb
 filesize=3M
 rw=write
 numjobs=40
+fsync=3
 
 [5_thread]
 stonewall
@@ -64,6 +66,7 @@ directory=gcs/5mb
 filesize=5M
 rw=write
 numjobs=40
+fsync=5
 
 [7_thread]
 stonewall
@@ -79,6 +82,7 @@ directory=gcs/50mb
 filesize=50M
 rw=write
 numjobs=40
+fsync=50
 
 [9_thread]
 stonewall
@@ -97,6 +101,7 @@ directory=gcs/256kb
 filesize=256k
 rw=randwrite
 numjobs=40
+fsync=16
 
 [11_thread]
 stonewall
@@ -113,6 +118,7 @@ directory=gcs/3mb
 filesize=3M
 rw=randwrite
 numjobs=40
+fsync=3
 
 [13_thread]
 stonewall
@@ -129,6 +135,7 @@ directory=gcs/5mb
 filesize=5M
 rw=randwrite
 numjobs=40
+fsync=5
 
 [15_thread]
 stonewall
@@ -145,3 +152,4 @@ directory=gcs/50mb
 filesize=50M
 rw=randwrite
 numjobs=40
+fsync=50