Is it possible using sendfile() system call to reduce cpu usage and increment download speed ? #460
Comments
|
The transfer speed bottleneck is multifaceted; specifically for dufs, it is primarily caused by the network, not the disk. 10M/s is the limit of your network, not the limitation of dufs. On my network, dufs can easily reach speeds of nearly 900M/s.
|
I accept to it, but for sure I have tested my network, it's not the bottleneck in my case, maybe I must apply more info as below. I have two nas, A is arm arch, B is x86 arch. case 0: run dufs on A nas, and run wget on B nas, the speed is about 10~20M/s, as I said in this issue case 1: run rsync on B nas, copy file from A, can reach to 25M/s and up. case 2: run dufs on B nas and download file to A nas, 70M/s and up Obviously my network is not so fast(only 1Gb), but A --> B and B --> A should be the same speed, and the test show different result. |
|
More info, may be useful. I am running dufs on A nas with docker, which is old (v19.03). If I run dufs directly, download speed can up to 30M/s, faster then rsync, much better... And the arm cpu is RK3328, old and weak. |
Specific Demand
I run dufs on a weak arm cpu NAS, it works very well, until I want to download a large file (10+G for example), wget only has 10M/s download speed. I use a ssd and benched the disk read speed with dd, 150M/s is shown, of couse the network is ok too.
I found dufs use 100% single core cpu, may be that is the reason?
Implement Suggestion
When I use nginx,
sendfile onworks very well, cause it skips the read/write function and use sendfile system call, so is it possible add this feature in dufs? I haven't read the source code yet, is it will make dufs more powerful?The text was updated successfully, but these errors were encountered: