Exactly Moacir, that is my point.

A proper Distributed FIlesystem should not rely on any type of RAID as it can make its own redundancy without having to rely on any underneath layer (look at CEPH). Using RAID may help with management and in certain scenarios to replace a faulty disk, but at a cost, not cheap by the way. That's why in terms of resourcing saving, if a replica 3 brings those issues mentioned it is much worth to have a small arbiter somewhere instead of wasting a significant amount of disk space.


On 08/08/2017 06:09, Moacir Ferreira wrote:


Let's see what people say... But this is what I understood Red Hat says is the best performance model. This is the main reason to open this discussion because as long as I can see, some of you in the community, do not agree.

But when I think about a "distributed file system", that can make any number of copies you want, it does not make sense using a RAIDed brick, what it makes sense is to use JBOD.


*From:* fernando.fredi...@upx.com.br <fernando.fredi...@upx.com.br> on behalf of FERNANDO FREDIANI <fernando.fredi...@upx.com>
*Sent:* Tuesday, August 8, 2017 3:08 AM
*To:* Moacir Ferreira
*Cc:* Colin Coe; users@ovirt.org
*Subject:* Re: [ovirt-users] Good practices
Moacir, I understand that if you do this type of configuration you will be severely impacted on storage performance, specially for writes. Even if you have a Hardware RAID Controller with Writeback cache you will have a significant performance penalty and may not fully use all the resources you mentioned you have.


2017-08-07 10:03 GMT-03:00 Moacir Ferreira <moacirferre...@hotmail.com <mailto:moacirferre...@hotmail.com>>:

    Hi Colin,

    Take a look on Devin's response. Also, read the doc he shared that
    gives some hints on how to deploy Gluster.

    It is more like that if you want high-performance you should have
    the bricks created as RAID (5 or 6) by the server's disk
    controller and them assemble a JBOD GlusterFS. The attached
    document is Gluster specific and not for oVirt. But at this point
    I think that having SSD will not be a plus as using the RAID
    controller Gluster will not be aware of the SSD. Regarding the OS,
    my idea is to have a RAID 1, made of 2 low cost HDDs, to install it.

    So far, based on the information received I should create a single
    RAID 5 or 6 on each server and then use this disk as a brick to
    create my Gluster cluster, made of 2 replicas + 1 arbiter. What is
    new for me is the detail that the arbiter does not need a lot of
    space as it only keeps meta data.

    Thanks for your response!


    *From:* Colin Coe <colin....@gmail.com <mailto:colin....@gmail.com>>
    *Sent:* Monday, August 7, 2017 12:41 PM

    *To:* Moacir Ferreira
    *Cc:* users@ovirt.org <mailto:users@ovirt.org>
    *Subject:* Re: [ovirt-users] Good practices

    I just thought that you'd do hardware RAID if you had the
    controller or JBOD if you didn't.  In hindsight, a server with
40Gbps NICs is pretty likely to have a hardware RAID controller. I've never done JBOD with hardware RAID. I think having a single
    gluster brick on hardware JBOD would be riskier than multiple
    bricks, each on a single disk, but thats not based on anything
    other than my prejudices.

    I thought gluster tiering was for the most frequently accessed
    files, in which case all the VMs disks would end up in the hot
    tier.  However, I have been wrong before...

    I just wanted to know where the OS was going as I didn't see it
    mentioned in the OP.  Normally, I'd have the OS on a RAID1 but in
    your case thats a lot of wasted disk.

    Honestly, I think Yaniv's answer was far better than my own and
    made the important point about having an arbiter.


    On Mon, Aug 7, 2017 at 5:56 PM, Moacir Ferreira
    <moacirferre...@hotmail.com <mailto:moacirferre...@hotmail.com>>

        Hi Colin,

        I am in Portugal, so sorry for this late response. It is quite
        confusing for me, please consider:

        *1*- *What if the RAID is done by the server's disk
        controller, not by software?

        2 -**For JBOD I am just using gdeploy to deploy it. However, I
        am not using the oVirt node GUI to do this.

        3 -**As the VM .qcow2 files are quite big, tiering would only
        help if made by an intelligent system that uses SSD for chunks
        of data not for the entire .qcow2 file. But I guess this is a
        problem everybody else has. So, Do you know how tiering works
        in Gluster?

        4 - I am putting the OS on the first disk. However, would you
        do differently?


        *From:* Colin Coe <colin....@gmail.com
        *Sent:* Monday, August 7, 2017 4:48 AM
        *To:* Moacir Ferreira
        *Cc:* users@ovirt.org <mailto:users@ovirt.org>
        *Subject:* Re: [ovirt-users] Good practices
        1) RAID5 may be a performance hit-

        2) I'd be inclined to do this as JBOD by creating a
        distributed disperse volume on each server.  Something like

        echo gluster volume create dispersevol disperse-data 5
        redundancy 2 \
        $(for SERVER in a b c; do for BRICK in $(seq 1 5); do echo -e
        "server${SERVER}:/brick/brick-${SERVER}${BRICK}/brick \c";
        done; done)

        3) I think the above.

        4) Gluster does support tiering, but IIRC you'd need the same
        number of SSD as spindle drives. There may be another way to
        use the SSD as a fast cache.

        Where are you putting the OS?

        Hope I understood the question...


        On Sun, Aug 6, 2017 at 10:49 PM, Moacir Ferreira
        <mailto:moacirferre...@hotmail.com>> wrote:

            I am willing to assemble a oVirt "pod", made of 3 servers,
            each with 2 CPU sockets of 12 cores, 256GB RAM, 7 HDD 10K,
            1 SSD. The idea is to use GlusterFS to provide HA for the
            VMs. The 3 servers have a dual 40Gb NIC and a dual 10Gb
            NIC. So my intention is to create a loop like a server
            triangle using the 40Gb NICs for virtualization files (VMs
            .qcow2) access and to move VMs around the pod (east /west
            traffic) while using the 10Gb interfaces for giving
            services to the outside world (north/south traffic).

            This said, my first question is: How should I deploy
            GlusterFS in such oVirt scenario? My questions are:

            1 - Should I create 3 RAID (i.e.: RAID 5), one on each
            oVirt node, and then create a GlusterFS using them?

            2 - Instead, should I create a JBOD array made of all
            server's disks?

            3 - What is the best Gluster configuration to provide for
            HA while not consuming too much disk space?

            4 - Does a oVirt hypervisor pod like I am planning to
            build, and the virtualization environment, benefits from
            tiering when using a SSD disk? And yes, will Gluster do it
            by default or I have to configure it to do so?

            At the bottom line, what is the good practice for using
            GlusterFS in small pods for enterprises?

            You opinion/feedback will be really appreciated!


            Users mailing list
            Users@ovirt.org <mailto:Users@ovirt.org>

    Users mailing list
    Users@ovirt.org <mailto:Users@ovirt.org>

Users mailing list

Reply via email to