ZFS & BTRFS Filesystems Analysis

Analysis, Pages 4 (935 words)

Views

2) Results: Postmark's tests were repeated three times and averaged. It showed that EXT4 performed well for file sizes greater than 1KB and BTRFS was best for file sizes lesser than 1KB. Bonnie++'s result indicated that EXT4 was best in all tests. The author made a remark that performance was not up to the bar as guest OS was not working in native conditions.

V. ANALYSIS FOR THE GROWING DEMAND OF STORAGE

A. Performance Evaluation of ZFS and LVM (with ext4) for Scalable Storage System

Performance of ZFS and LVM for scalable storage system was determined, with SATA and iSCSI as disk accessing protocols [5].

Don't use plagiarized sources. Get your custom essay on

“ ZFS & BTRFS Filesystems Analysis ”

Get custom paper

NEW! smart matching with writer

iSCSI (SCSI over the internet) is emerging strong as IP storage protocol, hence author has taken the protocol for experiments [9].

1) Experimental setup: The author had- single machine in case of SATA & two machines in case of iSCSI. It involved different disk configurations- single disk, a volume of two disk and a volume of three disks. IOZone tool tested file sizes from 64KB to 2GB.

The filesystems were remounted after every test to eliminate cache effect. Sequential read and write throughput were evaluated. The author considered Linux volume manager on directly attached SATA as baseline.

2) Results: The performance of LVM over SATA was better than baseline for file sizes smaller than system memory, even for three disks configuration. LVM over iSCSI performed well overall, but when file size was greater than system memory the throughput dropped even below baseline. ZFS over SATA poorly performed. ZFS over iSCSI performed well with increasing number of disks.

So, need at least 3 disks to achieve better performance than baseline. But increasing more will not necessarily help as the Ethernet speed will be a limitation because it will then involve multiple iSCSI sessions which costs more resources. ZFS performed well for read operation but only for file sizes smaller than 4MB. The author omitted results of 64KB-128KB file size as ZFS was interfered by cache.

B. Ext4, Btrfs, and the others

The Study of BTRFS and EXT4 filesystems was conducted keeping EXT3 and XFS for reference [7].

1) Experimental setup: Two different configurations- single machine and server. For single machine desktop with two core CPU and single SATA drive. Whereas for server, RAID array were setup. The experiment operated on FFSB flexible file system benchmark suite for server configuration. The tests carried out were: first, 16 threads create 1GB file using 4KB writes, second, Mail server workload where the filesystem got seeded with 1 million files spread across 1000 directories and third, creation of 1024 files of 100MB (each of 16 threads then write 5MB of data at random locations into these files, then randomly write, read, delete files). Three tests for single machine included: first writing 30 kernel trees, second, reading all files in directory with 64000 files and third, 100 threads syncing and writing 100 files.

2) Results: The study yielded that for the single machine, in writing 30 kernel trees, BTRFS took 10% lesser time than EXT4. While Reading all files in directory with 64000 files, XFS outperformed BTRFS quite close. And when 100 threads were syncing and writing 100 files BTRFS performed better. For server, as the 16 threads created 1GB file using 4KB writes, XFS performed very well, EXT4 performs better that BTRFS and EXT3 but performance of BTRFS with nodatacow option performs equally as default BTRFS. For the mail server test, EXT3 was a bit better than EXT4, then BTRFS and XFS. But here in contrast to the previous test, BTRFS with nodatacow option was noticeably better that default BTRFS. And for the third test EXT3 won, then XFS, BTRFS with nodatacow, EXT4 while BTRFS default option being the poor performers.

VI. CONCLUSION

ZFS and BTRFS both are easy to manage but BTRFS despite being younger has improved a lot over time. The authors have analyzed the performance of ZFS & BTRFS filesystems for storage configurations like- single disk, multiple disks, over LVM, over SATA & iSCSI disk accessing protocols, etc. Experimenting with different file sizes, record sizes, number of threads, and other mount parameters were opted to understand filesystem behaviour better. Majority of microbenchmark tests included sequential/random read/write, using IOzone tool. Mostly, BTRFS showed a higher throughput. Macro-benchmark tests included workloads of web server, mail server, OLTP, etc, using Filebench suite. For all tests of OLTP server, ZFS outperformed. Special mention has to be paper [4] where the author tested BTRFS especially on compression and defragmentation, captured the disk access behaviour using Guassian09, Blktrace and Seekwatcher. This study also asserted the effectiveness of BTRFS in-built defragmentation tool.

REFERENCES

[1] Mohd Bazli Ab Karim, Jing-Yuan Luke, Ming-Tat Wong, Pek-Yin Sian, Hong Ong, "Ext4, XFS, BtrFS and ZFS Linux File Systems on RADOS Block Devices (RBD): I/O Performance, Flexibility and Ease of Use Comparisons", 2016 IEEE Conference on Open Systems (ICOS), October 10-12, 2016, Langkawi, Malaysia.

[2] Anders Lundholm, "A Performance comparison of ZFS and BTRFS on Linux", Bachelor Degree Project at University of Skovde.

[3] Dj.Pesic, B.Djordjevic, V.Timcenko,"Competitionof VirtualizedExt4, Xfs and Btrfs Filesystems Under Type-2 Hypervisor", 24th Telecommunications forum TELFOR 2016.

[4] Meaza Taye Kebede, "Performance Comparison of Btrfs and Ext4 Filesystems", Masters thesis, University of Oslo, 2012

[5] Veerapat Phrornchana, Natawut Nupairoj, and Krerk Pirornsopa, "Performance Evaluation of ZFS and LVM (with ext4) for Scalable Storage System", 2011 Eighth International Joint Conference on Computer Science and Software Engineering (JCSSE).

[6] Dominique A. Heger, "Workload Dependent Performance Evaluation of the Btrfs and ZFS Filesystems", CMG International Conference 2009.

[7] JanKara,"Ext4,btrfs,and the others",InProceedingofLinux-Kongress and OpenSolaris Developer Conference, 2009.

[8] O. Rodeh, J. Bacik, & C. Mason, "BTRFS: The Linux B-tree Filesystem", ACM Transactions on Storage, Vol. 9, No. 3, Article 9, Publication date: August 2013.

[9] Jeffrey S.Goldner,"TheEmergenceofiSCSI", FileSystems and Storage Volume 1, issue 4, ACM Queue July 14, 2008.