help40 Posted May 23, 2016 Posted May 23, 2016 I want to make myself home NAS .... and looking for A20 board I found "Banana Pi M1+" $40 and "Banana Pro"$49 but I can not chose which is better ... , yes they have $9 difference in price ... but other seems the same , or there is other boards which I did not notice ? can you help me to chose what is better for home NAS (probably will install armbian and openvault ...) ?or other board suggestion in this price range I was looking for A20 because of SATA port ... yes there are other board with USB3 but they are in other price range than you in advance for your suggestions
tkaiser Posted May 23, 2016 Posted May 23, 2016 You should take an hour and read carefully through this article here: http://linux-sunxi.org/Sunxi_devices_as_NAS What differentiates Banana Pro from the original Banana Pi (or M1) is just onboard WiFi, same applies to M1+ that has been designed to be a Banana Pro clone (after the LeMaker and SinoVoip divorce -- of course SinoVoip failed with a few details so OS images for Pro are NOT 100 per cent compatible with M1+). In my personal opinion all these cheap onboard WiFi solutions aren't worth a look so I would neither choose Pro nor M1+ but the original Banana Pi instead (also cheaper). In case you want to use WiFi do some research before regarding your specific use case. And please remember that M1+ is made by a company 'famous' for not being able to provide correct hardware descriptions (be it .fex files or .dts files -- they still simply don't give a sh*t about correct information -- until this is resolved I would not buy anything from them). Regarding M1+ (and WiFi) also worth a read: http://forum.armbian.com/index.php/topic/340-banana-pi-pro-m1plus-no-wifi-armbian-45/ http://forum.armbian.com/index.php/topic/1149-patches-for-m1-dts-file-for-wifi/ http://forum.armbian.com/index.php/topic/1079-the-armbian-status-for-banana-pi-m1-plus/ BTW: We can expect an A20 successor called A20E rather soon (internally called sun8iw11p1, still dual core, still SATA capable but clocked somewhat higher). But since this chip seems to be based on different IP blocks compared to 4 years old A20 while it might be 100% pin compatible to A20 we currently know nothing what to expect regarding drivers and general compatibility). 1
help40 Posted May 23, 2016 Author Posted May 23, 2016 ....In my personal opinion all these cheap onboard WiFi solutions aren't worth a look so..... no ,in my case I do not need wifi at all .... 1000 mb network will be good my internet is "coming" from 1000 mb network port .... from provider GPON router which already has wifi build in (Huawei HG8245T) when you said "original banana pi" as I understand this is the same as "Banana Pi M1" ? like this one ? http://www.aliexpress.com/item/Free-Shipping-BPI-M1-Banana-Pi-A20-Dual-Core-1GB-RAM-Open-source-development-board-singel/32576359846.html so many modifications confuse me in my home I use wired network for my home computers .... wires everywhere so instead of using GB switch may be it is an idea to use Bpi R1 ? but I do not khow how performant is it in ... "simple switch mode" so may be "original banana pi" will be the best option
tkaiser Posted May 23, 2016 Posted May 23, 2016 The 'original' Banana Pi is called 'M1' when sold by SinoVoip. As far as I understand LeMaker is still selling this board just with the name 'Banana Pi'. Regarding your questions: Please be aware that you're talking about sunxi devices (Allwinner). So the only comprehensive information source is linux-sunxi wiki (again: take your time to read through the whole wiki pages): http://linux-sunxi.org/Banana_Pi http://linux-sunxi.org/Lamobo_R1(this board has so many design flaws that it should be avoided!) SinoVoip also tried to provide an informative chart with M1 and M1+ (A20 based so they are a good choice) and the totally incompatible M2, M2+ and M3 boards (only M2+ being a good choice since software support for the H3 SoC there is progressing): http://forum.banana-pi.org/t/all-banana-pi-sbc-comparison/1714 (as usual wrong since SinoVoip doesn't give a sh*t about correct information. Banana Pi can cope with SD cards larger than 32 GB and A83T as used on Banana Pi M3 has not a SGX544MP2 GPU but only SGX544MP1 -- half the count of GPU cores compared to the older A31s as used on BPi M2. They are not able to provide information/documentation that is correct). Depending on where you're living you might be able to get a LinkSprite pcDuino Nano 3 (Lite) for less money (Lite available in the US for just $15 sometimes ago)
Per-Mattias Nordkvist Posted May 24, 2016 Posted May 24, 2016 Hi guys, I'm jumping in to this thread. I really appreciate your input @tkaiser I have been playing around with _all_ the boards(raspberry, mostly Cubieboards(A10/20), Banan Pi, Orange Pi(Mini 2, PC, One, Lite)) and I have a couple of questions regarding performance. The SATA interface makes for really good write/read speeds but I am now considering using USB Flash Drive storage(with a RAID system) instead of Hard Drives. There are many advantages and therefore moved away from A10/A20 to H3 processors(due to pricing). Is it just my imagination or is the H3 boards a little bit slower than A20? On heavy duty like modifying video it performance fast but on the more regular tasks in the OS it seems lagging. Like just loggin in with multiple terminals to the same computer can be slow. Or when I try to perform multiple tasks at the same time. I read somewhere that A20/A10 is more of a real SoC where more is done internally in the chip. Is that true? You guys know more on this than I do.
tkaiser Posted May 24, 2016 Posted May 24, 2016 I read somewhere that A20/A10 is more of a real SoC where more is done internally in the chip. Is that true? Nope. A20 is dual core and has 1 SATA port, 2 USB host ports and one USB OTG. H3 has twice the CPU cores, can be clocked slightly higher (and will be then twice as fast if the workload is multithreaded!) and the SATA port is replace with another true USB host port. Some H3 boards expose all available USB host ports (Orange Pi PC, Orange Pi PC Plus, Orange Pi Plus 2E, NanoPi M1), some only a few (BPi M2+, OPi Lite, OPi One) and some use an internal USB hub to provide USB receptacles that have to share bandwidth (OPi 2, OPi Plus, OPi Plus 2). Regarding 'on the more regular tasks in the OS it seems lagging': This is mostly due to rootfs and/or user homes being stored on devices that show random I/O perfprmance that is way too low. What most people don't get: An SBC is not a digital camera or video recorder. Therefore the speed class of SD cards is close to irrelevant (sequential transfer speeds important for cameras/recordings) but random I/O matters. And there the differences are huge: http://forum.armbian.com/index.php/topic/954-sd-card-performance/(TL;DR: Buy Samsung EVO with at least 32GB at the moment since they're cheap, max out sequential transfer limitation of most Allwinner SoCs and show superiour random I/O performance) Same applies to technologies like SATA and USB: There we can also talk about random and sequential I/O. And while A20's SATA 2.0 implementation is somewhat limited regarding sequential write speeds (maxes out at ~45MB/s) SATA implements NCQ and shows superiour random I/O with fast SSDs. With USB it depends. Sequential transfer speeds max out at 35MB/s with BSP kernel or with mainline kernel and if the USB-to-SATA bridge used is UASP capable at 40MB/s. But more importantly random I/O benefits even more from UASP so in case you put rootfs or user homes on an USB disk it will be less sluggish when you can use UASP instead of the slower BOT mode (you're limited to when you have to use 3.4.x BSP kernels). So most likely what you're experiencing is just I/O performance and totally unrelated to the SoC in question. When you're running off an SD card with a recent Armbian install you could give armbianmonitor -c $HOME a try (will check the whole partition where $HOME resides and also measures sequential and random I/O performance and provides warnings if too low). In case you use a large and slow SD card this test can take ages so better have a look for the iozone call from the aforementioned 'SD card performance' thread.
help40 Posted May 24, 2016 Author Posted May 24, 2016 Oh, I did not know those things ! So in some cases board without "real SATA" but with 4.x.x kernel and UASP capable USB to SATA adapter will get almost the same speed (I am talking for HDD transfer speed) as board with SATA and even more if 2 adapters are plugged ... can use raid ... great
tkaiser Posted May 24, 2016 Posted May 24, 2016 Oh, I did not know those things ! Obviously you missed the first sentence I wrote in this thread? You should take an hour and read carefully through this article here: http://linux-sunxi.o..._devices_as_NAS What's the reason putting all this stuff into a wiki article? To not repeat it again and again partially in different forum posts. But since people in 2016 lack reading skills more and more when the amount of text exceeds a typical tweet... BTW: RAID as most people try to implement it with SBCs is crap anyway. But since people love RAID and hate backup they should get what they deserve: data loss/corruption
Per-Mattias Nordkvist Posted May 24, 2016 Posted May 24, 2016 This forum is one of the most relevant threads in this area. Thank your for great response! Do you mean that installing OS on USB flash drive would be better for Random I/O performance than running on a SD card?
tkaiser Posted May 24, 2016 Posted May 24, 2016 Do you mean that installing OS on USB flash drive would be better for Random I/O performance than running on a SD card? No. Since 'USB flash' is 'flash with USB controller' and SD card is 'flash with SDIO controller'. There are limitations by the host's interface (be it USB or SDIO) and there are controllers that manage the flash memory and there are the flash/NAND dies. You can get fast SD cards -- again see http://forum.armbian.com/index.php/topic/954-sd-card-performance/ -- and you can get slow USB flash stuff. And vice versa. So it depends. The most common mistake is that people don't want to understand what's important when we're talking about 'storage for the rootfs or user homes'. That's random I/O and not sequential transfer speeds. And most cards especially 'no names' and the like (PNY, Kingston, Verbatim and all the others that only buy NAND dies and controllers somewhere and assemble them to slow SD cards) really suck when it's about random I/O. So if you use 'the average SD card' you can buy at a shop then most probably yes moving the rootfs to USB flash storage might be way faster. But why speculating? Simply test it (Armbian ships with iozone so give the iozone call from the thread above a try). And using an A20 board together with a SSD that shows high IOPS values will outperform any USB 2.0 based solution and of course any SD card. When we're talking about rootfs and the appropriate use case (eg. trying to use an SBC as desktop replacement). In case I want to set up a headless server with data on a connected HDD (either SATA or USB) and with the rootfs still on SD card then with Armbian's defaults it won't matter that much whether the SD card is fast or slow since card accesses will only happen from time to time if the data to be shared is on the connected HDD. In the aforementioned article a few ways to identify bottlenecks are already mentioned. It's worth a look/try (eg. doing an 'apt-get install sysstat' and then running 'iostat 5' in a window, then 'doing normal stuff' that appears laggy and watch the iostat output for %iowait and %cpu)
Per-Mattias Nordkvist Posted May 24, 2016 Posted May 24, 2016 Great article. I will run more stress tests of my different options(I have probably 10 different vendors both with SD and USB SD). Thanks for all input.
Per-Mattias Nordkvist Posted May 26, 2016 Posted May 26, 2016 I've run more tests now and I am quite comfortable with the performance of the system now. Thanks for all input. I also noted that moving from bash to dash had a great improvement in the lagginess of working on the system. It reduces boot times as well.
tkaiser Posted May 26, 2016 Posted May 26, 2016 I also noted that moving from bash to dash had a great improvement in the lagginess of working on the system. It reduces boot times as well. The former sounds very unlikely and the latter close to impossible. Can you provide the output from 'sudo armbianmonitor -u' please?
Recommended Posts