I have no idea on the confusing sprawl that you are posting but my friend @LucidScrubJay apols if being over familiar, but really its not me.
The result for the 970 Evo was
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 70395 70811 92238 92886 46631 70865
102400 16 141961 200984 236280 237984 147557 204966
102400 512 500244 536614 466642 479526 467919 536983
102400 1024 529906 542864 483609 489514 483614 542280
102400 16384 652449 670618 628992 654369 644341 665665
That was the only bench on multiple file lengths as started that one as its common with with Armbian and on the internet.
Also its not known of that is gen1 or gen2 enabled but presuming gen 2.0
It might not make all that much difference as will use x4 lanes of gen 1.0 but its is gen 2.0 so enable it.
Not sure why its an overlay but hey I have forgot several times also.
Still seems odd to me that I need to edit a file to enable gen2 of the pci to get ther performance. If it’s backwards compatible why not default to gen 2?
Anyway, will have a look and see what happens
(ETA 13hours though)
Hi cevap,
Yes, I have run multiple 970 as raid0 too and above 2 combined you have almost saturated the PCI bus of 4x lanes, so usually it will be diminishing returns after that.
I realize that the 970 might be more than the rock can utilize on its pci bus, but I did not mind the prize as I can use the nvme on another project I have if the rock cannot get the most from it, and simply swap with one of my other nvme’s so that is fine for me still need to do some more testing before that though.
Keep in mind, I wanted fast IOPS and a fast responsive system, and do not need the larger capacity storage of the 4port expansion that you are testing.
For me the 1 nvme is enough.
I am still booting from SD until I settle with a distro I am happy with. (if majaro ends up supporting the nvme I will likely go with that. (have given up a little on using void for the rock)
Ignoring what the memory and CPU can deduct/contribute as bottleneck, if we look at the PCI throughput capabilities we can plan our storage accordingly.
Now in Radxa defense, likely not a lot of benchmarking had taken place before the 970 was available in the store, but looking at PCI Gen 2 4x lanes theoretic throughput should be close to 2.0 GB/s.
And I was okay if it just got over 1.0GB/s considering the CPU etc as I knew if that was the max capacity I had an old nvme I could swap it with instead.
Must admit that I would stay away from USB as I have seen USB drop devices quite often on normal pc’s and if you are running raid0 you will have data loss. Maybe okay if not for important data?
Or if you are running zfs raid that can self heal when dropping and adding USB devices.
okay, so it seems you have a lot of assumptions in here as well..
I agree that raid do not brake as often as people think, and especially when using a dedicated bus with the right connectors
USB is per definition already hot swappable.
I dont have these experiences with only winslows, linux and mac as well.
and if you boot and not all drives in your raid0 is detected you will have issues, if one of your disks drops during a write you will have issues, if the USB bus you are connected to is not dedicated you can have interupts, cachewrite and latency issues when you plug in or take out other equipment on that bus.
raid0 obviously have no redundancy and so if you data matters you only use raid0 for temp space while working on it and sawing to raid5/10. (and even in industry you would not take the risk of working on raid0, only for data you are prepared to lose, but for temp space it is okay.)
And yes it is always good to be able to fix your raid if there are issues, but in normal industry you will want to minimize the change of having to do so, hence most stay with dedicated BUS, or standard connectors for something like that.
However ! I think it is great fun to play around with and I think everyone should do whatever works for them. I never said you should not do it, just that I would not
In fact I think it is super interesting when someone tries these things and shares benchmarks etc.
Actually in comes from the standpoint of business continuity, return time objective and Return point objective when we valuate criticality for business processes. If a business process is critical and must resume after a very short time after stopping, the business will want the system architect to ensure that risks of recovery complications and instability are minimized, even at the cost of higher capital investment. And the company I am in have switched to VDI so on the serverside it is all NAS arrays so not really relevant for this discussion.
… but back to our talk of home users having fun with these single board computers as I think this thread is getting a little derailed haha.
I apologize if I was not really part of the beginning of this thread and sort of just got pulled in in the middle.
Look, I think you are busy trying to defend against arguments others have said ?
I never said you cannot or should not do it, heck I think its great that you tried and even documented and shared it with us
(and no I have not had a chance yet to see your scripts as using raid was not relevant for my use case, but sounds interesting, might have a look later)
Besides, there are few things more enjoyable than proving to people that say you cannot do it, that they are wrong haha.
WWell, years back I ran raid0 on Linux (as well as other platforms) and I have had data loss, and headaches with it. That was why I did not want to introduce more of the same headaches using USB in the first place.
However, this was quite a few years back so might be fun to try again sometime. For now at least, it is not needed for me though.
Going back to home users having fun or having simple systems then this is where USB drives can come in really handy and also for many of us there are some cheaper options.
I really like snapraid and unionfs and if you are running Openmediavault it makes these utilities even more fun, simple is as simple does.
It is so easy and failure of any disk is again so easy, you just get a collection of any size disks make sure your parity is the biggest and go.
Then just grow and add disks when needed and you can use this guide to when you need to add a new parity disk.
Snapraid is remote raid of a collection of random disks without any need for constant sync timings and its creates parity (something to rebuild from) on schedule which for home use maybe daily, with a movie store it might even be weekly.
As generally the downtime is acceptable.
Then unionfs just puts them together as a volume and writes to the disk with the most space.
Simples innit and perfect for home and perfect for USB.
If you where cunning enough to get a high wattage 12v PSU with barrel adaptor then its just a matter of splitter cables and one of these and hubs.
You can go completely crazy and stick another 4 port USB3.0 into the M.2 and you can fan out to a huge store that also has simple redundancy and often even relatively slow 3.5" 7200 rpm disks are still maxing out what is common of a 1gbe network.
USB does have its uses.
Tried and tested but afterwards thinking hold on that is only 1 lane but hey who needs math!
Load up your benches @LucidScrubJay when you get back if you can as interesting if gen1.0 to 2.0 will make any difference.
Unlike some its the small numbers that are of most interest.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
102400 4 70395 70811 92238 92886 46631 70865
The 4k rec length is highly indicative of server and client processes and its only us that like to lumber around with huge files, but generally its the smaller files that have the most impact on feel and performance.
Will be interesting to see if there is any improvement.
yep, will do
big file benchmarks are good for fileserver performance when getting ready for a NAS setup, but yeah application, db servers and snappy response in linux is more about the small IOPS.
I do suspect that for the smaller file transfers it wont change much, but I hope I am wrong haha.
Will find out later.
Even for file servers its only really media servers / home servers where singular large files have sustained throughput as with higher concurrency its chopped into many concurrent packets.
It doesn’t take all that many for the effect to be felt and those small file values quickly start becoming the main performance factor.
Even on a client desktop where cores and parallelism is becoming ever more its not really about singular file transfer and tests such as these and higher job counts are more indicative of many modern multi process situations.
And..you know, when majority of devs with experience in theme they are talking about says “No, that’s bad choice”, MAYBE it means it’s actually bad choice
But why would you use Raid0??? It’s risky (it IS, there is now way around this, unless you are going for some crazy tricks which doesn’t work unless you use specific OS), it’s doesn’t bring anything performance wise since we are still limited to 1 gb/s internet speed, which roughly 128mb/s. And most solo HDD already have that speed.
I couldn’t care less for any of this? Why would you ping me?
Would be great if you could td;lr all of this… Would be much more simply. Also, why in the hell USB testing being done within “PCIE-to-Sata adapter” thread?
That’s…wow. Please wash your face next time before typing anything on forums please. The only not smart person i see there is you, if that how you answer to question about why there so much off-topic coming from you.
And that’s still not answer why you started this conversation there. In Sata benchmark’s thread. Nor this answer any of my questions. And if you thinks it’s fine to post off-topic everywhere…Well… I guess you are worst kind of person there.
