Date: Thu, 4 Jan 2018 18:23:30 -0800 From: Mark Millard <markmi@dsl-only.net> To: Andreas Schwarz <freebsd.asc@strcmp.org> Cc: Freebsd-arm <freebsd-arm@freebsd.org> Subject: Re: Times to build just devel/llvm50 via poudriere-devel: Pine64+ 2GB, RPi3, RPi2 V1.1 Message-ID: <C2E42018-A753-40B4-93C7-6769122FAD4E@dsl-only.net> In-Reply-To: <4b428a608dd.290862fc@mail.schwarzes.net> References: <65D7B16B-E3D7-40F2-BE60-0EE5E5B26B31@dsl-only.net> <4b428a608dd.290862fc@mail.schwarzes.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2018-Jan-4, at 3:40 PM, Andreas Schwarz <freebsd.asc at strcmp.org> wrote: > On 04.01.18, Mark Millard wrote: > >> eMCC performance notes: >> The rpi3 can get between 10 MiByte/s and >> 12 MiByte/s, while the Pine64+ 2GB can get >> between 5 MiBytes/s and 6 MiBytes/s, from >> what I have observed. > > Unfortunatly the DiskIO (to SD Card) of the Pine64 is very slow > in comparsion to RPI2B (or RPI3). > > root@rpi2b:~ # dd if=/dev/zero of=/var/zero.img bs=1m count=1024 > 1024+0 records in > 1024+0 records out > 1073741824 bytes transferred in 80.823823 secs (13284967 bytes/sec) > > root@pine64plus:~ # dd if=/dev/zero of=/var/zero.img bs=1m count=1024 > 1024+0 records in > 1024+0 records out > 1073741824 bytes transferred in 383.600722 secs (2799113 bytes/sec) > > Both devices running r327391 with without any debugging, INVARIANTS, > etc.. Used SD Cards: Samsung MB-MGBGB SDHC 32GB > > Mark, can you perform the write test with your eMMC to MicroSD > Adapter? [I probably should have noted that I mount with -o noatime for all these contexts. And all 3 have heat sinks and fans. The RPi2B V1.1 and Pine64+ 2GB were running -r327364 at the time (and still are).] The RPi2 context is odd because the usdcard slot is partially broken: Its mechanism for hold in cards has failed but its mechanism for ejecting them still works. This is one reason why there is a USB SSD stick on a powered USB hub involved: USB is where the kernel, the dtb file that the kernel reads, the root file system, and the swap space are all from. I hold in the usdcard until the kernel starts to load then I let go. As the poudriere bulk activity was based (RPi2B V1.1: USB SSD stick). . . RPi2B V1.1 (cortex-a7) using the USB SSD stick: # dd if=/dev/zero of=/var/zero.img bs=1m count=1024 1024+0 records in 1024+0 records out 1073741824 bytes transferred in 48.251071 secs (22253223 bytes/sec) (Note: The powered hub and USB SSD stick are all USB 3.0 capable and for a good USB 3.0 context can sustain well over 10 times the above figure.) RPi3: (cortex-a53) using a 64 GB eMMC via a usdcard adapter: # dd if=/dev/zero of=/var/zero.img bs=1m count=1024 1024+0 records in 1024+0 records out 1073741824 bytes transferred in 72.442738 secs (14821939 bytes/sec) Pine64+ 2GB: (cortex-a53) using a 128 GB eMMC via a usdcard adapter: # dd if=/dev/zero of=/var/zero.img bs=1m count=1024 1024+0 records in 1024+0 records out 1073741824 bytes transferred in 189.560701 secs (5664369 bytes/sec) The RPi2B V1.1 and RPi3 have similar amounts of RAM, and so likely similar amounts of swapping activity (compared to the Pine64+ 2GB), but the slower I/O rate context (RPi3) took less time for the build than the faster I/O rate context (RPi2B V1.1). The Pine64+ 2GB did some swapping, but not much compared the the RPi2/3 (far more RAM). With this difference (and hardware differences), the slowest I/O rate of the 3 contexts took the least build time. RPi2B V1.1 eMMC via an adapter: no figures, I'm afraid. . . Each of my attempts to hold an eMMC with an adapter in the RPi2 usdcard slot for a test have resulted in a panic. I'm afraid I'll not be getting an RPi2B V1.1 timing for an eMMC via an usdcard slot adapter. eMMC adapter note: Some cases for single board computers interfere with using an eMMC via an adapter. === Mark Millard markmi at dsl-only.net
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?C2E42018-A753-40B4-93C7-6769122FAD4E>