linux-kernel - fadvise+sync_file_range Vs O

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <88df6dda0704030351u44fab6bcj20f9b48d24de861a@mail.gmail.com>
Date:	Tue, 3 Apr 2007 16:21:21 +0530
From:	"Ameya Mitragotri" <ameya.mitragotri@...il.com>
To:	linux-kernel@...r.kernel.org
Cc:	suparna@...ibm.com, pbadari@...il.com
Subject: fadvise+sync_file_range Vs O_DIRECT

Hello,
I am doing some benchmarking on writes to the hdd from a simple
application (attached).
Im benchmarking the writes using the following
- write with O_DIRECT
- write with fadvise + O_SYNC
- write with fadvise + sync_file_range

Ive tested the writes in chunks of 4/8/16 KB and 1MB

The results show fadvise + sync_file_range is on par or better than
O_DIRECT. Detailed results are attached.

Now are these results along the expected lines? AFAIK O_DIRECT should
be faster because it bypasses the cache. Does it make sense for apps
to use fadvise + sync_file_range rather than use O_DIRECT?

I havent tried benchmarking using tools like Iozone as yet.

Hardware/OS details
1) AMD Athlon with 512 MB RAM running plain vanilla 2.6.20 (ext3 partition)
2) Intel P4 with 512 MB RAM running plain vanilla 2.6.20 (ext3 partition)

Thanks
Ameya

Download attachment "appl.tar.gz" of type "application/x-gzip" (5498 bytes)

View attachment "odirect.txt" of type "text/plain" (1902 bytes)

View attachment "fadvise+o_sync.txt" of type "text/plain" (2056 bytes)

View attachment "sync_file_range+fadvise.txt" of type "text/plain" (1918 bytes)