See https://www.flexense.com/usb3_vs_sata_disk_performance_comparison.html Here local SATA3 vs USB3 comparison is done. While not directly comparable, the only case where throughput is below what you can push through GbE is single threaded small file copying. For every other case tested, GbE will be the bottleneck.
> On Sep 21, 2019, at 5:32 AM, hiro <23h...@gmail.com> wrote: > > yeah, but check small blocksize random read/write vs. AoE or 9p over > ethernet. I'm not sure how efficient usb3 in terms of latency :) > > On 9/21/19, Bakul Shah <ba...@bitblocks.com> wrote: >> On Fri, 20 Sep 2019 09:53:07 +0100 Richard Miller <9f...@hamnavoe.com> >> wrote: >>> >>>> Another option worth exploring may >>>> be AOE as pi4 has a GbE (I haven't tried this yet). >>> >>> My go test builders are running with "local" fossil on a slice >>> of disk provided over AoE from an atom server. I tried various >>> configurations and this gave me the best performance. This is >>> with 3B+ machines on "gigabit" ethernet throttled by rubbish usb. >>> >>> Pi4 has proper GbE, but also has usb3 so a local ssd drive might >>> be a practical alternative. More experiments to do. >> >> On linux/pi4 I get about 230MB/s for seq. read on a $10 USB3.1 >> Samsung flash drive. Time to get a new SSD! >> >> >
