Hi Guys or Girls,
I have a problem in a productive system. We use three Servers which have
a Raid 5 with four Harddrives SATA 8 TB 7200 rpms. The Systems are
interconnected for glusterfs with 1 GB/s.
The Network Load between all three servers are between 5 and 8 % of the
network. The Glusterfs has a separate VLAN which compltete isolated from
other networks.
Find attached a small drawing.
On that server we rum as HA and there are 8 VMs installed. Most of the
VMs working well but if there is high load on the system some of the VM
can't write the Data in a "good" speed on the disks. The Glusterfs is
configured as Replica on the Raid5 disk which is as Jbob in Glusterfs
defiend.
If i take a look on the iostats i can see that the Harddisk have round
50 open connections some of the glusterfs and a lot of the vms. In side
the VMs i can see long iowaiting times from 300 up to 600 ms. With is on
case of a Posgresql database very long.
Should we configure the Raid5 controller out that the controller only
should 4 disks with 8 TB and Glusterfs should connect directly to the
/dev/sd[b-e]?
What do you think? Will SSD HDD will help on performance but the same
config with Raid5 and Jbod could be the bigger problem.
Thanks
Marcel