TOPIC:

Hello,
i think about a classic 2 Node Setup.

I want to export 2 iSCSI Targets for the virtual SAN managed by HW-Raid:
1: 6 x NL-SAS HDD`s for mostly static Data
2: 2 x SATA SSD`s for DomU OS`s / DB-Server and similar

iSCISI / DRBD will sync on 2x 10GbE.
(I`m aware that i need to adjust rates in the DRBD configs ... )

First question is about RAID-Level:
i wonder if RAID 10 might be over-redundant and if it`s not sufficient to have a RAID 0 on each node as the cluster itself is sort of a RAID 1.

Second is about the SSD`s / TRIM:
I found some posts that SSD Write performance can suffer due to the lack of TRIM - Support in Xenserver.
Will this even be an issue with HW-Raid in each node ?
Does anyone have a number what throughput can be expected in a DomU on a SSD-backed Target in such a Setup ?

Third: are there any issues running Dom0 on a M.2 SSD or SATA-DOM besides SPOF of these ?

Last but not least: any suggestions what Raid-Cards to use ?
2 Port internal + 2 Port external for future expansion as 'nice to have' + BBU is what i`m looking for.

And, is support for Dundee on the Roadmap ?

Thanks,
Alex

Thank you for your post.
Regarding whether RAID 10 is over redundant: it really depends on the application and how much down time one is willing to accept in the event of a HW failure. On the surface, replicated RAID 10 provides 4x replication of data and seems over redundant - but consider the case where a single disk partition (non-RAID) fails on a dom0 that is running VMs. In this case the dom0 host would continue to run the VMs, but, the VMs would all lose thier backing disk. In this case there is nothing to trigger an HA recovery since the VMs and dom0 are still running.

In the same scenario with replicated RAID 10 - a disk failure would have NO impact on running VMs and if the cluster were properly managed it could alert an admin of the failed disk which could be replaced while operational. The result is no downtime at all and no risk of data corruption.

Regarding your questions about SSDs and suggested RAID controllers. All of our testing and development (and live deployments) is done on HP servers using SAS drives (no SSD) so no real comment on how SSD would perform. You may get more input on that toipic on the Citrix/XenServer official forum.

Lastly - regarding Dundee - the short answer is yes. We do plan on supporting the next major release of XenServer but have not yet started the process of testing/patching to work aginst the centos7 style dom0.

Hi, thank you for your answer.

consider the case where a single disk partition (non-RAID) fails on a dom0 that is running [...] In this case there is nothing to trigger an HA recovery

Is this also the case for a RAID 0 (what, in theory, could be configured with a single drive on the raid-card) that is synced by DRBD and managed by HA-iSCSI ?
I think in this case the whole SR would fail on the host if served from local, what would be a good reason to switch SR`s iSCSI target to the host with the intact Disks. Then the failed disks could be replaced and after a DRBD resync the host should be fine again.

If not, what would happen then ?
For example, let a RAID 5 on the Host with the iSCIS Target IP crash because a 2nd disk fails while in resync.

Correct - it would be the same scenario for RAID 0 and a RAID 5 that loses 2 disks. Currently iscsi-ha does not detect disk failures. Perhaps it is something that we can look at in the future. In the meantime, I would avoid non-RAIDed disks.

Wow, ok ...
Does this really mean that in the case of a storage subsystem failure, even when managed by HA-iSCSI, there will be a serious problem / no failover of the iSCSI ip to the other host ?

I was sure the active ip of the SR would be managed, too.