Re: measuring swap partition speed

Warner Losh <imp_at_bsdimp.com> wrote on
Date: Fri, 15 Dec 2023 15:41:10 UTC :

I'm trying to figure out what to do with the
fio, what adjustments to the example are
appropriate, both for the orginal message's
context and for what I have access to.

> . . .
> Now, having said all that, your best bet is to run a FIO test. fio is my
> go-to choice for doing
> benchmarking of storage. Do a random workload with a 8k write size (since
> that's the page
> size of aarch64)

I'm guessing the figure in question is:

# sysctl vm.stats.vm.v_page_size
vm.stats.vm.v_page_size: 4096

Is that correct?

> on one of the swap partitions when it's not in active use.
> I suspect you have
> a SSD, and that it will kinda suck, but be in line with the swap
> performance you are seeing.
> 
> I use the following template for my testing (128k should be reduced to 8k
> for this test,

Again:

# sysctl vm.stats.vm.v_page_size
vm.stats.vm.v_page_size: 4096

> though
> I've not looked at how much we cluster writes in our swap code, so maybe
> that's too pessimistic).

Is that based on:

# sysctl vm.phys_pager_cluster
vm.phys_pager_cluster: 1024

I'm unclear if vm.phys_pager_cluster is relevant or,
if it is, how vm.phys_pager_cluster should be put to
use here.

Where is the 8k figure from?

> You might also try reducing the number of I/O jobs, since I'm measuring, or
> trying to, what the
> best possible sustained throughput numbers are (latency in this test tends
> to run kinda high).

The original E-mail reported: rpi4 and a usb3-connected disk

So: 4 cores and the USB3 is one channel that the 2 USB3
ports share. (There are 2 separate USB2 ports.)

Later "spinning rust" was mentioned for the media. (I've
only USB3 SSD or USB3 NVMe access.)

> ; SSD testing: 128k I/O 64 jobs 32 deep queue

Spinning rust, not an SSD for the original message.
(But I've no access to spinning rust to test:
USB3 SSD or USB3 NVMe access.)

Again: 128k's replacement?

64 jobs and 32 deep: might not be a good match to
4 cores and single channel USB3 spinning rust? Or
my context: 4 cores and USB3 SSD or USB3 NVMe?

> [global]
> direct=1
> rw=randread

Only reads? I'd have guessed a mix of reads and writes,
not necessarily in fixed proportions. So I'm not sure
if rw=randwrite was the intended replacement or if
something else that would end up exploring mixes would
be the intended replacement.

> refill_buffers
> norandommap
> randrepeat=0
> bs=128k

4k? 8k? something else?

> ioengine=posixaio
> iodepth=32

???

> numjobs=64

???

> runtime=60
> group_reporting
> thread
> 
> [ssd128k]

===
Mark Millard
marklmi at yahoo.com