linux-ext4 - Append and fsync performance in ext4 DAX

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]

Date:   Sat, 28 Apr 2018 11:24:32 -0500
From:   Vijay Chidambaram <vvijay03@...il.com>
To:     Ext4 <linux-ext4@...r.kernel.org>,
        Rohan Kadekodi <kadekodirohan@...il.com>,
        aasheesh kolli <aasheesh.kolli@...il.com>
Subject: Append and fsync performance in ext4 DAX

Hi,

We have been running some performance experiments on ext4 DAX and
noticed some behavior that is puzzling us. We'd be grateful if someone
could explain what is going on.

Workload 1:

create a file
while file size <= 5GB:
   append 4K
   fsync
close file

Running time: 20 s

Workload 2:

open 5GB file
for each block in file:
   overwrite 4K
   fsync
close file

Running time: 1.8 s

While we expect workload 1 to take more time than workload 2 since it
is extending the file, 10x higher time seems suspicious. If we remove
the fsync in workload 1, the running time drops to 3s. If we remove
the fsync in workload 2, the running time is around the same (1.5s).

We initially suspected this was because of delayed allocation. But
when we mounted ext4 without delayed allocation, we got similar
results.

We ran this on ext4 with -o dax option on /dev/pmem0, where the data
on a write() system call is written using non-temporal writes.
However, we also ran this on ext4 without dax on top of ramdisk
(/dev/ram0), and found no difference in the numbers. So we believe the
root cause to be in ext4 and not in /dev/pmem or /dev/ram device.

Could someone help figure this out?

Thanks,
Vijay