3 3
Igor

research Marvell based 4 ports mPCI SATA 3.0

Recommended Posts

Wonderful how the net churns - gotta love the wayback machine. Poindexter would be proud.

 

Still, I would like to do some performace testing - what do you suggest?

https://docs.armbian.com/User-Guide_Fine-Tuning/#how-to-provide-boot-logs-for-inspection

is helpful. I've also used screen's ctrl-a : hardcopy -h filename to save displayed text.

 

But what are you running to get the speed stats? I've found instructions for many different ways, but I'd like to be 'on the same page' with you on these kinds of tests. You always say 'testing testing testing testing...' so that's what I'm doing, just want to be able to compare apples to apples with you guys.

 

Thanks again

Share this post


Link to post
Share on other sites
On 16.2.2018 at 7:32 PM, Technicavolous said:

But what are you running to get the speed stats?

 

You need a bunch of terminal/SSH windows and then do as root

  • armbianmonitor -m (to get an idea whether throttling happens, how %iowait looks like, how cpufreq scaling is working)
  • htop (to see whether the benchmark you're running is bottlenecked by one or more CPU cores being at 100% -- if that's the case you found a CPU bottleneck and need to find ways around that)
  • iostat 5 (to get some storage statistics)
  • iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2 (the real test)

If you see in htop CPU cores being maxed out at 80% or above you need to check /proc/interrupts for IRQ collissions and might have to pin the benchmark execution itself to another CPU core, e.g. using taskset -c 1 iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2 to send the test to the 2nd CPU core called cpu1.

 

And then you need a 'baseline' when testing with storage (eg. a beefy x86 box with native SATA where you benchmark the storage device before to get an idea about it's real capabilities) since otherwise you have no idea where to look at when numbers are lower than expected (is it the board bottlenecking or the storage device?)

 

Unfortunately there's also something called software and since Armbian constantly tries to be bleeding edge with bootloaders and kernels it got 100% stupid to collect here any benchmark numbers since the performance influence by different kernel versions can be MASSIVE. As an example: If you choose for an Clearfog the Debian Stretch next image from here https://dl.armbian.com/clearfogpro/ then SATA performance will be SHIT until you run 'apt upgrade ; reboot' since with latest 4.14.18 kernel available through Armbian repo performance is good again. So benchmarks made on 1st February look crappy while those made today with latest kernel show an entirely different picture :( 

Share this post


Link to post
Share on other sites

That is excellent information - I can only compare once I know what I have ... and sincerely, if it is 'good enough' that works for me. I just need to figure out where I am and have a consistent way of doing so because things do indeed change as you said.

 

I don't have a 'beefy' x86 machine but I have a decent one. That's a good idea to compare against.

 

Thanks again for the good information. 

Share this post


Link to post
Share on other sites
3 3