In the last 10 years, Ext4 has risen in prominence, ReiserFS has fallen to the wayside, ZFS has been ported to Linux, XFS keeps plugging along, and there's a new kid: Btrfs. NTFS has evolved, too. It's now 2016. How do these filesystems stack up against each other? Does it really make that much of a difference? We’ll show you the results of standard, consistent tests across platforms (Linux vs. Windows) and filesystems to see if the differences are worth choosing one over the other. For simplicity's sake, the tests are performed on identical hardware with out-of-the-box settings.
2. 2
The Problem
Perforce on Solaris is sunsetting
Standardizing on Linux
• Which filesystem?
Previous Merge presentations a decade old
Perforce has recommendations, no numbers
3. 3
The Methodology
Standard Perforce branch-submit benchmark
Test on same physical hardware (retired master)
Use out-of-the-box settings (aside from compression)
Include Windows
Software RAID drives (lvm, ZFS, BTRFS, Windows volume)
Run tests five times, chart average
4. 4
The System
Master server retired in March 2015
• 2 physical processors (Intel Xeon 2.2 Ghz)
• 16 cores (8 per CPU)
• 512 GB memory
Ten SAS drives
• 3 TB each
• Samsung 7200 RPM 6 Gbps
P4D 2015.1
5. 5
Filesystems and OS’s
Solaris 11 x86_64
• ZFS (uncompressed/compressed lzjb)
RedHat Linux 7.2
• ZFS (uncompressed/compressed lzjb)
• Ext3/Ext4
• Xfs
• BTRFS (uncompressed/compressed lzo/compressed zlib)
Windows Server 2012R2
• NTFS/ReFS
19. 19
Final Thoughts
Windows performs generally worse when compared to
almost any Linux filesystem.
XFS is a solid choice. Supported, mature, fast.
BTRFS compressed vs uncompressed not hugely different.
Ext3 holds up surprisingly well. Go figure.
20. 20
Questions For You
Are ZFS features worth the performance hit? (likely)
Is BTRFS ready for primetime? (perhaps)
Is Windows performance poor enough to justify introducing
Linux into your data center? (yes, yes it is)
21. Contact us for raw data
or further questions.
majanu@dolby.com
sjo@dolby.com