Re: [Openstack-operators] ceph vs gluster for block

[Alex Hübner]

Gluster for block storage is definitely not a good choice, specially for
VMs and OpenStack in general. Also, there are rumors all over the place
that RedHat will start to "phase out" Gluster in favor of CephFS, the "last
frontier" of the so-called "Unicorn Storage" (Ceph does everything). But
when it comes to block, there's no better choice than Ceph for every-single
scenario I could think off.

[]'s
Hubner

On Thu, Feb 16, 2017 at 4:39 PM, Mike Smith  wrote:

> Same experience here.  Gluster ‘failover’ time was an issue for as well
> (rebooting one of the Gluster nodes caused unacceptable locking/timeout for
> a period of time).  Ceph has worked well for us for both nova-ephemeral and
> cinder volume as well as Glance.  Just make sure you stay well ahead of
> running low on disk space!  You never want to run low on a Ceph cluster
> because it will write lock until you add more disk/OSDs
>
> Mike Smith
> Lead Cloud Systems Architect
> Overstock.com 
>
>
>
> On Feb 16, 2017, at 11:30 AM, Jonathan Abdiel Gonzalez Valdebenito <
> jonathan.abd...@gmail.com> wrote:
>
> Hi Vahric!
>
> We tested GlusterFS a few years ago and the latency was high, poors IOPs
> and every node with a high cpu usage, well that was a few years ago.
>
> We ended up after lot of tests using fio with Ceph cluster, so my advice
> it's use Ceph Cluster without doubts
>
> Regards,
>
> On Thu, Feb 16, 2017 at 1:32 PM Vahric Muhtaryan 
> wrote:
>
>> Hello All ,
>>
>> For a long time we are testing Ceph and today we also want to test
>> GlusterFS
>>
>> Interesting thing is maybe with single client we can not get IOPS what we
>> get from ceph cluster . (From ceph getting max 35 K IOPS for % 100 random
>> write and gluster gave us 15-17K  )
>> But interesting thing when add additional client to test its get same
>> IOPS with first client means overall performance is doubled  , couldn’t
>> test more client but also interesting things is glusterfs do not use/eat
>> CPU like Ceph , a few percent of CPU is used.
>>
>> I would like to ask with Openstack , anybody use GlusterFS for instance
>> workload ?
>> Anybody used both of them in production and can compare ? Or share
>> experience ?
>>
>> Regards
>> Vahric Muhtaryan
>> ___
>> OpenStack-operators mailing list
>> OpenStack-operators@lists.openstack.org
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>
> ___
> OpenStack-operators mailing list
> OpenStack-operators@lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>
>
>
> ___
> OpenStack-operators mailing list
> OpenStack-operators@lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>
>
___
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

Re: [Openstack-operators] ceph vs gluster for block

[Mike Smith]

Same experience here.  Gluster ‘failover’ time was an issue for as well 
(rebooting one of the Gluster nodes caused unacceptable locking/timeout for a 
period of time).  Ceph has worked well for us for both nova-ephemeral and 
cinder volume as well as Glance.  Just make sure you stay well ahead of running 
low on disk space!  You never want to run low on a Ceph cluster because it will 
write lock until you add more disk/OSDs

Mike Smith
Lead Cloud Systems Architect
Overstock.com



On Feb 16, 2017, at 11:30 AM, Jonathan Abdiel Gonzalez Valdebenito 
> wrote:

Hi Vahric!

We tested GlusterFS a few years ago and the latency was high, poors IOPs and 
every node with a high cpu usage, well that was a few years ago.

We ended up after lot of tests using fio with Ceph cluster, so my advice it's 
use Ceph Cluster without doubts

Regards,

On Thu, Feb 16, 2017 at 1:32 PM Vahric Muhtaryan 
> wrote:
Hello All ,

For a long time we are testing Ceph and today we also want to test GlusterFS

Interesting thing is maybe with single client we can not get IOPS what we get 
from ceph cluster . (From ceph getting max 35 K IOPS for % 100 random write and 
gluster gave us 15-17K  )
But interesting thing when add additional client to test its get same IOPS with 
first client means overall performance is doubled  , couldn’t test more client 
but also interesting things is glusterfs do not use/eat CPU like Ceph , a few 
percent of CPU is used.

I would like to ask with Openstack , anybody use GlusterFS for instance 
workload ?
Anybody used both of them in production and can compare ? Or share experience ?

Regards
Vahric Muhtaryan
___
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
___
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

___
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators

[Openstack-operators] ceph vs gluster for block

[Vahric Muhtaryan]

Hello All , 

For a long time we are testing Ceph and today we also want to test GlusterFS

Interesting thing is maybe with single client we can not get IOPS what we
get from ceph cluster . (From ceph getting max 35 K IOPS for % 100 random
write and gluster gave us 15-17K  )
But interesting thing when add additional client to test its get same IOPS
with first client means overall performance is doubled  , couldn¹t test more
client but also interesting things is glusterfs do not use/eat CPU like Ceph
, a few percent of CPU is used.

I would like to ask with Openstack , anybody use GlusterFS for instance
workload ? 
Anybody used both of them in production and can compare ? Or share
experience ? 

Regards
Vahric Muhtaryan


___
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators