• DevSummit
• 201510
• Performance

Performance testing

Chair Person: DanLangille

Abstract:

There are many things we can measure with performance testing. The chair has particular areas of interest but welcomes others to join with their own topics.

The following list started with the chair's topics, but was expanded as others joined in.

• Opening: Areas of performance to test
• ZFS resilver rates
  • There are many myths about resilver rates but few data regarding relative speeds
  • e.g. a 10 disk raidz3 versus a 10 disk raidz2, what is the resilver time?
  • at 25% full, 50%, 75%, 85%?
  • when the system is idle, when under load?
  • how does 10x2.5TB drives compare to 5 x 5TB drives?
  • there are many ideas and suggestions, but nothing comparing configurations on a relative basis

Attending

In order to attend you need register for the developer summit as well as by email for the session and be confirmed by the working group organizer. Follow the guidelines described on the main page or what you received by email. For questions or if in doubt ask the session chairs.

Please do NOT add yourself here. Your name will appear automatically once you have received the confirmation email. You need to put your name on the general developer summit attendees list though.

Notes

The original notes taken during session: https://etherpad.net/p/2015Dev_Summit_Performance

Copy paste of above, so it's here:

Matt Ahrens Blog about ZFS stripe width: http://blog.delphix.com/matt/2014/06/06/zfs-stripe-width/

How long does it take to resilver with:

• differing numbers of drives
• used space (25%, 50%, 75%, 85%)
• different raid levels
• different fragmentation levels
• resilver times under load, and while idle. (how to simulate an even level of load, mbuffer -R)

raid z: repeat benchmarks for each number of disks, and look for *-modal distributions

bcr: steal his dtrace script steal Adam Leventhal's dtrace scripts, and Brenden Greggs jmg has latency checking script

Ensure block sizes of the drives are identical (expect 4k) (For enterprise, most of the HDD are still 512B of block size (512n or 512e?)) 512 native.

marcelo / bapt have some benchmarking tools: Zopkio to dispatch, collect and process logs (Used at gandi)

zfs snapshot performance, delete 10,000s of snapshots = degraded performance until reboot benchmark 'zfs list' with different levels of snapshots

disable during benchmarking:

• crontab
• sendmail

Use an automated framework like this: https://github.com/ocochard/netbenches https://github.com/ocochard/netbenches

compression benchmarks

• lz4 vs gzip
• gzip 1 - 9