Analyzing new Truenas Core 13 System performance

[JDE1000]

Hello all. New Truenas Core 13 user and also fairly new to BSD/Linux. I apologies up front for any newbieish questions.

I have the new system in my signature and I have run a bunch of tests and it seems like my pool speeds are not in line.

Running Iperf3 between my Truenas system and a 10GB workstation via a 10GB switch I get approx 9.4 GB.

Transferring files between the same systems (32 GB file) I get 330 MB write and 220 read

Fio results for the command "fio --name=random-write --ioengine=posixaio --rw=randwrite --bs=64k --size=256m --numjobs=16 --iodepth=16 --runtime=60 --time_based --end_fsync=1" are:


Run status group 0 (all jobs):
WRITE: bw=154MiB/s (161MB/s), 8718KiB/s-11.5MiB/s (8927kB/s-12.0MB/s), io=13.5GiB (14.5GB), run=89945-89950msec

Before I changed to Truenas Core I used WIn 11 and got 210-220 MB seq read/write on Crystaldiskmark

Are these results normal? Can I improve these results?

Thank you

 

[morganL]

3 SATA drives in a RAIDZ1???

 

[JDE1000]

3 SATA drives in a RAIDZ1???

Sorry. All 4 drives are in a stripped pool. I'm not interested in redundancy

 

[morganL]

My guess is the USB drive is the slowest...and causes imbalance. If you are just testing, you might want to test without it.

A wagon train only goes as fast as the slowest wagon.

 

[MrGuvernment]

USB drive in a pool = bad idea
L2Arc = likely completely useless and could cause slower performance (More memory should always come before adding a L2ARC drive

 

[JDE1000]

USB drive in a pool = bad idea
L2Arc = likely completely useless and could cause slower performance (More memory should always come before adding a L2ARC drive

Thank you. my MB is limited to 64 GB. I average 54 GB ARC. How can I determine if the L2Arc is helping or hurting.

 

[morganL]

L2ARC rarely helps unless you have many clients accessing the same data or some other very specific use-case.
Its almost pointless testing without a real workload.

 

[JDE1000]

L2ARC rarely helps unless you have many clients accessing the same data or some other very specific use-case.
Its almost pointless testing without a real workload.

Thank you. I use the system almost exclusively for torrenting. 3 different clients.