[Gluster-users] fuse vs libgfapi LIO performances comparison: how to make tests?

[luca giacomelli]

Dear support, I need to export gluster volume with LIO for a 
virtualization system. In this moment I have a very basic test 
configuration: 2x HP 380 G7(2 * Intel X5670 (Six core @ 2,93GHz), 72GB 
ram, hd RAID10 6xsas 10krpm, lan Intel X540 T2 10GB) directly 
interconnected. Gluster configuration is replica 2. OS is Fedora 27

For my tests I used dd and I found strange results. Apparently the 
volume mounted locally and exported with LIO is more faster than the 
same volume exported directly with LIO.

For my tests I exported the volume from the first server and mounted and 
checked the volume on the second

This is the dd command:

echo 3 > /proc/sys/vm/drop_caches

dd if=/dev/urandom of=file_1G bs=1M count=1000 oflag=direct

with fuse

1073741824 bytes (1.1 GB, 1.0 GiB) copied, 9.5597 s, 112 MB/s

direct with user:glfs

1073741824 bytes (1.1 GB, 1.0 GiB) copied, 27.0812 s, 39.6 MB/s

I think there is something strange. I don't really know if dd is good in 
order to obtain "good" numbers. I read that some people use iozone other 
Fio or iometer...

Before ask information i'd like to be sure that the procedure and the 
instrument used to check the configuration is correct.

I made also other tests and all suggest to me that the fuse way is 
better than the other. Is it possible?

Any help would be appreciated.

Thanks, Luca