apt-get iozone3
iozone -a -e -I -i 0 -i 1 -i 2 -s 80M -r 4k
Gives a relatively quick and good test without it going on for ages.
Dunno you will have to test a2/a1
Think the SD 3.0 reader doesn’t take advantage of Micro SD express so the min guaranteed IOPs doesn’t happen.
Think Class A1 is the fastest the reader will take and use to full extent.
Nice, good to know. Will definitely try this tonight 
thanks for the hint
Not sure what I am doing wrong, but the results I get for the 64GB A2 uSD card does not seem indicative of the speed I am seeing when just farting around, and booting. Also cannot seem to declare me m.2 as a target for the test, when using the "iozone /dev/nvme0n1 --> iozone /u10 I get the same speed results as if using root. 
just mount the m.2 cd to the mount dir and run as it runs from current dir
lsblk to see your disks
mkdir m2
mount /dev/xxx m2
cd m2
iozone -a -e -I -i 0 -i 1 -i 2 -s 80M -r 4k
PS if that was your A2 yeah pretty bad my a1 is faster.
thanks, I am obviously too ired at this point haha.
I did mount the m2 but thought I should just pass the mount point for the test to run there instead of moving to the mount point. Will try that next 
And yes it is super slow in this test, but that seems wrong, the system is very responsive and very quick to boot. the Sandisk extreme should not be that slow.
Trying out and just running update for the Manjaro KDE version now
So far its quite slow compared to the openbox that runs on the ARM64 but seems very “polished” and complete very nice.
Will let it update completely and reboot and see how it fairs
Not surpricingly the m2 disks comes up with better numbers in the test.
Samsung SSD 250GB 970 EVO Plus M.2
PS Sandisk Pro 32gb A1
Command line used: iozone -a -e -I -i 0 -i 1 -i 2 -s 80M -r 4k
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
81920 4 4763 4971 8408 8400 6711 4916
But got you beat
Command line used: iozone -a -e -I -i 0 -i 1 -i 2 -s 80M -r 4k
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
81920 4 128739 332329 601247 615462 578960 336148
awesome numbers
what is the 2nd round results from ?
m.2 brand and make?
Lols, I couldn’t beat that evo 970 so I cheated.
As well as adding some zram swap which works well on the Rockpi4, I cheated and added a zram dir and ran from there.
zram dir make really good zero block wear exceptional fast application directories if you ever need to.
Also you can create a zlog in a small 50M zram dir and just stop avoid block wear there is you so wish.
Still waiting for my 4 port sata but doubt I will be anywhere near your evo plus
haha fair enough, you did have me scratch my head there for a sec 
And yes thats a pretty sweet (and oldschool) way to do speedup as long as it is for non important data.
Cool, I did see the post about the 4 port sata haha thats then starting to look like a pretty cool NAS setup if you do zfs and pool those sata with zfs cache running in that zram
This looks really cool, will definitely have a play with this cheers man.
Ram disk is as old as I am, but back then memory cost the tip of a fighter jet, so might be cool to dedicate a little of the spare ram most of us have these days for something like this.
I also have an idea to play around with some of the cluster possibilities of the SBC’s (not trying to create anything crazy, more for the sake of messing around and learning something in the process. my attention is fleeting and short lived and my time is very sparse haha but I like to try new stuff when time permits. it always a tradeoff, I used to have plenty of time-ish, and so I messed around with unix and gentoo linux but now that work takes most of my time (macOS and Winslows) its hard to find the time … /rant
Anyways, my idea which is currently “on the back of a napkin after a drunk night out” kinda note, is to try and do some of the distributed computing that I have seen others do with raspberry pi.
But have the hardware on some of the newer 1gbe NIC SBC’s and have them connected over websocket for ultra fast latency, and distribute commands by mqtt with pub/sub for communications. (think if all are subscribing to a master node that can send RPC calls over mqtt to all at the same time) and maybe have each be part of a dynamic zpool owned/controlled by the masternode with dynamically adding/removing of nodes of the sbc you attach. their contribution in storage could then be from a zram mounted disk (and yes there is a good chance someone have already done this)
Okay, as you have like noted by now, this idea is super crazy and full of holes that I have not even been close to consider yet, but I like the idea. And maybe only some of this is possible, but hey might be a fun project.
Anyways this is WAY off topic, so sorry about that, just thought that something like the Rock Pi with its 6cores could be a good candidate for a project like this in the future, and using your zram might be a cool addition?
zram-config is just a protest project about some very bad scripts but zram is pretty good also in 5.1 lzo-rle for arm is supposed to be amazingly fast whilst getting x3 compression.
The alg is your choice with zlib/zstd having great text compression up x10 but slower/more cpu.
The 2gb RockPi4 is prob best value for money and with zram if you change swappiness and page cluster its like what the do in chromiumos / android. 50% ram disksize x3 pagecluster 0 swappiness 100
But also like said if you do have more ram 4gb you can also make extremely fast working directories.
I use overlayfs to use CoW to bring the original data in and so only writes are copied up to zram so it can use large directories in minimal ram whilst on shutdown merges down to keep persistent.
Really needs a more modern kernel as a lot of overlayfs fixes happened but it works fine for most especially logs and swap doesn’t use overlayfs.
Your evo 970 is so fast that its almost near ram speed but the iozone gives you an indication.
I am humming and harring about ZFS with 4 disks and wondering if ZRAID is really much of an advantage over MDADM but going to test it out and see.
Just trying to get OMV working fine on 4.4 maybe also with Nextclould and LibreOffice Online in a docker container with Samba using a 4 disk raid.
The 1gb ethernet on the Rockpi4 is excellent but via USB or gadget g_ether its actually pretty stinky about 60% of the dedicated 1gb nic port.
If you struggle making your own void prob Manjaro minimal with i3 on 5.2 will be just as good if not better…
yeah, actually I turned an old macbook air mid-2011 into a Manjaro with awesome WM and it just screams in use so might be a good fit in the future. not decided yet, but will have it running for a while and evaluate.
For the zfs, I think the advantages are pretty overwhelming, in terms of snapshot, copy-on-write, self healing, dynamic raid etc etc. at the cost of a few cpu cycles more though. But if it is a dedicated NAS those cpu cycles are likely not needed for anything else anyway.
Dunno yet
PS a slight update to the image if you want to have a quick look.
OMV Arm64 Image
https://1drv.ms/u/s!AocmAh35i26QiRoVLn1ttnlorw8g
Debian user:password rock:rock
OMV user:password admin:openmediavault
sudo systemctl enable resize-helper.service reboot to resize /
PS Lucid the 4 port sata turned up today the below is a single 250gb Evo 850 sata
Command line used: iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
102400 4 40059 51432 60812 63636 31464 50149
102400 16 106962 142357 160286 165256 95699 139100
102400 512 325343 335507 317536 323487 312129 338324
102400 1024 340292 351986 340852 339575 337433 352246
102400 16384 444705 452061 479606 482634 479726 454479
Sweet then 4 of those in raid5 ?
If meant for OMV or NAS I guess the extra speed won’t matter as your 1GBe bandwidth is saturated already but the redundancy -and if zpool - self healing fs properties will be nice.
Yeah just testing that they will and how it handles multiple disks prob a stripe, 5 & 10 just for benches.
But really the dest is HDD but if it can handle these on expected benches then should pretty much expect HDD max.
Its a shame the g_ether gadget transfer is a bit stinky forgot what it was and guess will have to try that again.
What do you get from your Evo spacecruiser with iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2 ?
Will give it a run tonight when home from work and let you know
Btw, also ordered the 16gb emms and usb3 key. But if I can boot from m.2 after flashing I guess that will be moot?
Still haven’t worked out if you need spi nor or not if you have 1.3 version then its a solder job as its missing.
But the Manjaro guys seem to think you can do it without via uboot, but how slightly confused.
But answer is yeah you can its sort of new and needs a fix but answer is yeah.