ZFS a UFS vs ZFS benchmark

W

weberjn

I could try FreeBSD 14.1 on a KVM-based virtual server with 8 dedicated CPUS from an EPYC 9634 and 16G Ram, 1TB on a hardware raid and ran some benchmarks. I am posting the results, somebody might be interested.

I ran the Dacapo Java benchmark that uses some well known Java applications.

I installed once with UFS on Geli, ran the benchmarks and then reinstalled with ZFS on Geli (by the installer) and ran the benchmarks again. It is somewhat slower on ZFS with a much higher variance (so I have added more than one result of the same benchmark).

Code: 
java -version
openjdk version "17.0.12" 2024-07-16

$ java -jar dacapo-23.11-chopin.jar -C lusearch

UFS

===== DaCapo 23.11-chopin lusearch PASSED in 4044 msec =====
===== DaCapo 23.11-chopin eclipse PASSED in 11488 msec =====
===== DaCapo 23.11-chopin tradesoap PASSED in 3527 msec =====
===== DaCapo 23.11-chopin tradebeans PASSED in 5209 msec =====
===== DaCapo 23.11-chopin kafka PASSED in 5359 msec =====
===== DaCapo 23.11-chopin xalan PASSED in 1030 msec =====

ZFS

===== DaCapo 23.11-chopin lusearch PASSED in 4304 msec =====
===== DaCapo 23.11-chopin lusearch PASSED in 4460 msec =====
===== DaCapo 23.11-chopin eclipse PASSED in 11669 msec =====
===== DaCapo 23.11-chopin tradesoap PASSED in 3705 msec =====
===== DaCapo 23.11-chopin tradesoap PASSED in 3564 msec =====
===== DaCapo 23.11-chopin tradebeans PASSED in 5239 msec =====
===== DaCapo 23.11-chopin kafka PASSED in 5232 msec =====
===== DaCapo 23.11-chopin kafka PASSED in 5356 msec =====
===== DaCapo 23.11-chopin xalan PASSED in 1831 msec =====
===== DaCapo 23.11-chopin xalan PASSED in 1946 msec =====
===== DaCapo 23.11-chopin xalan PASSED in 1539 msec =====
 
You can use benchmarks/fio for benchmark.
When you are testing it must be done on bare hardware not on VM where you hit the cache buffers of the hypervisor or cache of the raid controller so you can't actually measure the performance.
What you can test is hardware Raid10 with BWCC with UFS against same disks in HBA with ZFS with striped mirrors.
 
VladiBG said:
You can use benchmarks/fio for benchmark.
When you are testing it must be done on bare hardware not on VM where you hit the cache buffers of the hypervisor or cache of the raid controller so you can't actually measure the performance.
Click to expand...

A benchmark on bare hardware does not answer the question, which filesystem is faster in a virtual server.
This question arose from this post about ZFS on virtualized disks.
 
I would still be curious what the host is and what filesystem the VMs are hosted on.

And the same benchmarks on the host.
 
The benchmark that he's running is CPU intensive not HDD.
1727242507089.png
 
You must log in or register to reply here.
Back
Top