Actually there are tons of possible tune ups, but let's start from the
beginning.
Gluster documents, Red Hat documents, and several other sources suggest
using XFS, and NOT ext4.
Second, How did you measure performance ? What tool ?
>From my personal experience, first of all, measure performance of you
hardware, before measuring Gluster's.
I found out a major problem with my disk layout. I was naive enough believe
that by using a good RAID controller (PERC 6) with fair disks (SAS 7.2 KRPM
3 Gbits) on RAID5 I would be in the safe side. It was a disaster. It never
passed 40 MBytes/s.
Changing that to RAID10 made it 10 times faster, at the price of half of
the storage space.
Test your hardware.
Start with simple stuff like
dd if=/dev/zero of=filename_on_your_local_disks bs=512k count=10000
check what the result is, and if your hardware is responding as expected.
Only then move to other areas.
Cheers,
Carlos
On Tue, Apr 29, 2014 at 10:47 AM, Dragon <Sunghost at gmx.de> wrote:
> Hello,
>
> i have a 3 Node Gluster with one distributed Volume. All 3 Nodes are
> identically in their Hardware. All have 6x3TB Seagate ST3000DM001. The disk
> on each Node are in a Softwareraid6 (Blocksize 4096| Stripe 512) with ext4
> filesystem on Debian Wheezy. As the Raid was new i testet it with fio:
> [sequential-write]
> rw=write
> size=5g
> direct=0
> numjobs=3
> group_reporting
> name=sequential-write-buffered
> bs=4k
> runtime=10
>
> Result: WRITE: io=3014.2MB, aggrb=308456KB/s, minb=308456KB/s,
> maxb=308456KB/s,
>
> I added a new sata controller Digitus DS-30104-1 4 Port PCIe x4, each Port
> 6GBits on x16 Slot. Node1 has now 7 Disks in Raid6. I added another Disk
> 3TB same model too on the Controller and copied files 20-40GB onto the
> volume. The Volume is mounted on the same node with "mount -t
gluster-fs
> IP-Adress:/vol1" The Problem is that the writespeed is only 20MB/s.
This
> is what i didnt understand. The files are copied on the same machine, not
> over the network, perhaps the bus-system of the mainboard is to slow? Any
> kind of performance tuning?
>
> thx
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140429/420a01c8/attachment.html>