Why slows down glusterFS?

Question

linux

Good afternoon.
Set up a glaster between two servers in replication mode. Each server will connect to itself as a client.
On top of this lives samba with balls of these folders.
There are about 200 thousand small files in the folders (10-20kb each) ...
The download speed through samba is about 20kb, the upload speed seems to be quite good .. (did not notice any problems). The speed of opening folders through samba is terrible.
sometimes input\output errors occur when localhost reads from glaster's folder... (rarely)
PS samba with regular fs (not glaster) works with a bang.
Please tell me, what could be the problem with this behavior of the glaster?
Tested without tuning and with tuning:
gluster volume set vol nfs.enable-ino32 on
gluster volume set vol nfs.addr-namelookup off
gluster volume set vol nfs.disable on
gluster volume set vol performance.cache-max-file-size 10MB
gluster volume set vol performance.cache-refresh-timeout 5
gluster volume set vol performance. write-behind-window-size 10MB
gluster volume set vol performance.io-thread-count 64
gluster volume set vol performance.cache-size 5G
Thanks
UPD1:
Torn a little rsync write test here (don't ask me why... wanted =) )
#### SSD TO SSD
sent 7672190769 bytes received 797076 bytes 180540890.47 bytes/sec
total size is 7668399515 speedup is 1.00
#### SSD TO GLUSTER ( on ssd ) with FUSE mount REPLICATED 2 blocks on two servers
sent 7672133481 bytes received 796791 bytes 18737314.46 bytes/sec
total size is 7668343095 speedup is 1.00
#### SSD TO GLUSTER ( on ssd ) with NFS mount REPLICATED 2 blocks on two servers
sent 7672401667 bytes received 799223 bytes 19207010.99 bytes/sec
total size is 7668603097 speedup is 1.00
#### SSD TO GLUSTER ( on ssd ) with FUSE DEPRICATED-REPLICATED of 4 blocks on two servers
speed is the same as and in the case of replicated.
Optics between glasterservers.
Somehow these results surprise me very much ... I came across articles on the Internet where people claim that their glaster downloads at the grid speed (1GB in their case was) ....

Answer 1

everything is working properly there.
This is the specifics
1. you do not have ssd and IO sags from the number of files and not from real work with them.
2. glusterfs works through FUSE and therefore it will not be fast in principle, together with the first point, it is really not fast.

Answer 2

ethernet? Servers in different DCs? If yes, then gluster is not for you.

Answer 3

GlusterFS is a mega-enterprise.
If you have one data center, then use Ceph as the base storage for the file system.
Better yet, Ceph + Object Storage.
If you have several data centers, then use Yandex Elliptics
https://habrahabr.ru/company/yandex/blog/228389/