Bananian

Pro & M2 Tuning

27 4632
skraw  
Your critics on using scp means one should not use owncloud with https either, right? Is that server-stuff in your eyes?

tkaiser  
Edited by tkaiser at Tue May 26, 2015 04:05

Neither hdparm nor scp are valid benchmarks if your use case is 'storage and network'. And the only reason I still write in this thread is because others might believe in your weird 'benchmarks' and conclusions

You can play your 'cached read' stuff and you could also measure ambient temperature of the rooms surrounding the machines you test (or your blood pressure while you're testing). The numbers you get are extraneous to anything related to network/disks. I tried to point you to some ressources here in the forums where you can get an idea of real world performance of SBCs in NAS scenarios (with and without tuning -- exactly the stuff you pretend to be interested in). Using an appropriate test setup one is able to achieve approx. 35/70 MB/s write/read to a Banana Pi through network including uncached disk access (it's that slow due to some GMAC and SATA limitations. Both might be lifted when DMA capable drivers are available/improved). This is simply impossible with any RPi since its single USB connection between SoC and the outside world is the bottleneck. No matter how you adjust core, RAM or any other timings. And no matter which unrelated stuff you measure.

This could've been a 3 post thread:

"Hey guys, why do I get so slow scp transfers when I use Bananian?"
"That's due to its hardened SSH configuration only allowing strong SSH ciphers which let one single CPU core serving the ssh process in question become the bottleneck. Use something different than scp/ssh in a LAN scenario when you care about performance. If you still want scp with Raspbian performance have a look at its sshd_config and weaken Bananian's config accordingly"
"Thx"



BTW: I do this scp/ssh stuff for a living. Even Intel CPUs from today aren't able to sature GBit LAN when dealing with strong SSH ciphers if they can't rely on AES-NI. So you have to use many rsync runs in parallel and weak ciphers like arcfour when you want to saturate the network and it's obvious that scp is heavily CPU bound and it's also obvious that it's the wrong tool to draw any conclusions about network or disk performance from. The performance you get when using scp depends heavily on the strength of encryption dynamically negotiated between both partners.

And BTW: I still don't care about your cached read stuff since it's still irrelevant for the use case you're always talking about

Even an completely untuned Banana Pi is 4 to 8 times faster in NAS useage scenarios than any RPi can be.

tkaiser  
Edited by tkaiser at Tue May 26, 2015 03:38
skraw replied at Tue May 26, 2015 03:26
Your critics on using scp means one should not use owncloud with https either, right? Is that server-stuff in your eyes?


If you're a programmer why don't you get that the performance differences you measure depend on software/settings instead of the hardware used? And if you're after CPU bound applications and the server has to serve many parallel sessions then you should better go with more CPU cores. If the server just serves one connection then RPi and BPi will perform nearly identical if settings are also identical. And if I would use ownCloud (which is not the case regarding the horrible security problems this software has) then I would not rely on slow SBCs but would at least use a x86-Micro-ITX board with Intel Celeron N3050 due to AES-NI. You do not get strong encryption as well as good performance for free.

From my side end of thread now. It's just a waste of time.

skraw replied at Tue May 26, 2015 10:26
Your critics on using scp means one should not use owncloud with https either, right? Is that server ...

Well, sure you can use Owncloud with https. But for most people, the network throughput won't matter much in that case because their upstream internet connection would be way slower - so if they want to access their owncloud, hosted on a Raspberry/Banana Pi at home, remotely while they're on the go, they wouldn't even get close to 100mbits or 1gbits bandwith (more like <10mbits for most people). For home use or local access, samba is probably the most widely used and preferred file server option (or alternatives like nfs, afp or ftp) which will allow for much faster transfer rates.

That being said - I'm out of this discussion. First, I wasn't really sure whether this thread is merely a troll attempt or whether it's just based on misconceptions/lack of knowledge but deserves answers. However, your attitude is beyond me, so I won't waste any more time here. You may go on to waste your's...

skraw  
Well, it makes no real difference if you write answers as they are pure praying that bananas are just better without a word about the very obvious problems the _software_ does seem to have. At least we are at the point where - additional to my complaint about badly configured ram and core speeds - you came up unveiling another problem of slow crypto implementation. The first you try to ignore, the second you suggest not having speedy internet connections, yet again ignoring that there may well be usage scenarios not involving internet at all.
If you don't have answers, please don't waste your and my time.
All I was asking for was an equally working patch for something everyone can solve on RPI(2) within 2 minutes of googling...

tkaiser  
Edited by tkaiser at Wed May 27, 2015 04:08

Wow, what a bunch of weird stuff... if you're speaking of 'slow crypto implementation' then it seems you don't understand crypto at all since scp/ssh or encryption in general is heavily CPU bound when there are no crypto accelerators available. And a distro like Bananian that enforces strong SSH ciphers will automagically lead to slower transfer rates. It's such a no-brainer to get this coherency!

What you call 'badly configured ram and core speeds' are just sane defaults. And you still fail totally to understand that the performance numbers you got are not the result of your 'tuning' of some clock speeds but due to the different contents of sshd_config (you even fail to understand how simple it is to verify this). Also the conclusions from available DRAM bandwidth to network/disk performance are just insane. And also your weird assumptions about the interdependence of DRAM to CPU clock. If you want to know details (which you clearly won't) you might want to read: http://linux-sunxi.org/DDR_Calibration vs. http://elinux.org/index.php?title=RPiconfig&section=14#Overclocking.

But this clock stuff is totally unrelated to the performance differences you've seen and we've been trying to point you to the real causes of the performance differences you experienced but since you still just want to believe the differences are the results of your magic 'working patch' (haha, a 'patch' -- do you know raspi-config?) it's just a waste of time to answer. But I did it one last time so that others stumbling accross this thread won't believe the insane assumptions you wrote.

Since feeding a troll is no good idea I summarized the stuff where it belongs to: http://forum.lemaker.org/forum.php?mod=viewthread&tid=15518

DrTune  
I'm with tkaiser here; I just got a BPi and it's really pretty good; vastly better performing than the original RPi; both Pi and Pi2 suffer (terribly imho) from the chronic bandwidth limitations of a single USB port shared for everything. I just managed to get 90MBYTEs/sec over the BPi's ethernet port using FTP - nearly saturating a gigabit link (and I wasn't even using jumbo frames - although this was data cached in the BPi's ram).  It's not perfect but it's damn good. For my application I'm very delighted to have _dual_ USB HS host ports (separate controllers; not through a hub), plus SATA, plus Gig-E, all on separate buses.  It's not beefy enough to saturate them all at once but it's a great board all around.

pandora jewelry
ugg outlet
longchamp handbags
columbia shoes
hermens bags
nike huarache
coach outlet
nike outlet
pandora
birkenstock outlet
fitflops sale
adidas shoes
lebron james shoes
michael kors
coach outlet online
ecco shoes
jimmy choo outlet
timberland boots
canada goose outlet
asics outlet
pandora jewelry
ysl handbags
jordan shoes
nike outlet online
valentino shoes
kate spade outlet
nike tessen
red bottoms
hermes bags
fred perry polo
christian louboutin outlet
dr martens outlet
adidas yeezy boost
cheap mlb jerseys china
adidas nmd
ralph lauren outlet
hermes outlet
polo ralph lauren outlet online
nike free 4.0
pandora jewelry
michael kors outlet clearance
coach outlet
coach outlet
christian louboutin
louis vuitton outlet store
adidas nmd shoes
birkenstock outlet
adidas nmd
birkenstocks
christian louboutin shoes
cartier outlet
coach outlet online
pandora jewelry
coach outlet store
ray ban sunglasses
yeezy 350 boost
michael kors outlet online
fred perry
louboutin shoes
michael kors outlet
supreme new york
kate spade outlet
jordan shoes
fitflops
ralph lauren sale clearance
michael kors outlet
cheap oakley sunglasses
pandora charms
uggs outlet
ralph lauren
jordan shoes
nmd adidas
kate spade outlet store
nike shoes
prada handbags
yeezy boost
pandora charms
cheap jordans
louboutin shoes
coach outlet
pandora jewelry outlet
michael kors outlet clearance
nike dunks
michael kors outlet canada
timberland outlet
fitflops sale clearance
pandora outlet
canada goose outlet
michael kors outlet canada
ralph lauren sale clearance
nike air max 1
red bottom shoes
oakley sunglasses
kobe bryant shoes
timberland boots
canada goose
air max 2018
adidas yeezy
kate spade outlet
toms outlet
180718yueqin

You have to log in before you can reply Login | Sign Up

Points Rules