On January 22, 2020 6:46:39 PM GMT+02:00, Benedetto Vassallo <benedetto.vassallo(a)unipa.it&gt; wrote: > Hello Guys, > Here at University of Palermo (Italy) we are planning to switch from > vmware to ovirt using the hyperconverged solution. > Our design is a 6 nodes cluster, each node with this configuration: > > - 1x Dell PowerEdge R7425 server; > - 2x AMD EPYC 7301 Processor; > - 512GB of RAM (8x 64GB LRDIMM, 2666MT/s, Quad Rank); > - 2x Broadcom 57412 Dual Port 10Gb SFP+ ethernet card; > - 3x 600GB 10K RPM SAS for the OS (Raid1 + hotspare); > - 5x 1.2TB 10K RPM SAS for the hosted storage domain (Raid5 + > hotspare); > - 11x 2.4TB 10KRPM SAS for the vm data domain (Raid6 + hotspare); > - 4x 960GB SSD SAS for an additional SSD storage domain (Raid5 + > hotspare); > > Is this configuration supported or I have to change something? > Thank you and Best Regards. > -- > Benedetto Vassallo > Responsabile U.O. Sviluppo e manutenzione dei sistemi > Sistema Informativo di Ateneo > Università degli studi di Palermo > > Phone: +3909123860056 > Fax:     +3909123860880 Hi, Recently it was mentioned that there were some issues with the 'too new' EPYC. For now, you can do : 1. Use some old machines for initial setup of the HostedEngine VM (disable all Spectre/Meltdown in advance) -> and then add the new EPYC-based hosts and remove the older systems. Sadly, the older  systems cannot be too old :) 2. Host the HostedEngine VM on your current VmWare environment or on a separate KVM host. Hosting the HostedEngine on bare metal is also OK 3. Wait (I don't know how long) till EPYC issues are solved. Best Regards,Strahil Nikolov

Hi Benedetto, we have a running cluster with the same machine and a similar configuration, currently we haven't encounter any issue. We're running ovirt 4.3.7.   Greeting,     Paolo Il 22/01/20 17:46, Benedetto Vassallo ha scritto: >  

Def. Quota Strahil Nikolov <hunter86_bg(a)yahoo.com&gt;: > On January 22, 2020 6:46:39 PM GMT+02:00, Benedetto Vassallo > <benedetto.vassallo(a)unipa.it&gt; wrote: >> Hello Guys, >> Here at University of Palermo (Italy) we are planning to switch from >> vmware to ovirt using the hyperconverged solution. >> Our design is a 6 nodes cluster, each node with this configuration: >> >> - 1x Dell PowerEdge R7425 server; >> - 2x AMD EPYC 7301 Processor; >> - 512GB of RAM (8x 64GB LRDIMM, 2666MT/s, Quad Rank); >> - 2x Broadcom 57412 Dual Port 10Gb SFP+ ethernet card; >> - 3x 600GB 10K RPM SAS for the OS (Raid1 + hotspare); >> - 5x 1.2TB 10K RPM SAS for the hosted storage domain (Raid5 + >> hotspare); >> - 11x 2.4TB 10KRPM SAS for the vm data domain (Raid6 + hotspare); >> - 4x 960GB SSD SAS for an additional SSD storage domain (Raid5 + >> hotspare); >> >> Is this configuration supported or I have to change something? >> Thank you and Best Regards. >> -- >> Benedetto Vassallo >> Responsabile U.O. Sviluppo e manutenzione dei sistemi >> Sistema Informativo di Ateneo >> Università degli studi di Palermo >> >> Phone: +3909123860056 >> Fax:     +3909123860880 > > Hi, > > Recently it was mentioned that there were some issues with the 'too > new' EPYC. > For now, you can do : > 1. Use some old machines for initial setup of the HostedEngine VM > (disable all Spectre/Meltdown in advance) -> and then add the new > EPYC-based hosts and remove the older systems. Sadly, the older  > systems cannot be too old :) > > 2. Host the HostedEngine VM on your current VmWare environment or on > a separate KVM host. Hosting the HostedEngine on bare metal is also OK > > 3. Wait (I don't know how long) till EPYC issues are solved. > > Best Regards,Strahil Nikolov Thank you. Maybe it's better to use Intel processors? Best Regards. -- Benedetto Vassallo Responsabile U.O. Sviluppo e manutenzione dei sistemi Sistema Informativo di Ateneo Università degli studi di Palermo Phone: +3909123860056 Fax:     +3909123860880

Hyperconverged uses gluster, and gluster uses replication (replica 3 or replica 2 + arbiter) so adding raid below may not be needed.

You may use the SSDs for lvm cache for the gluster setup.

I would try to ask on Gluster mailing list about this.