Anyone Using HPE Alletra 5010 in Production Environment

X25

New Member
Nov 13, 2026
4
0
1
Hi,

Looking for some feedback from anyone using HPE Alletra 5010’s for production workloads with Proxmox, and if you are any nuances to be aware of in configuring them to work with proxmox

We have a pair of 5010 and will be looking at a clustered proxmox setup running a virtual machines

Thanks in advance
 
One thing I'd check before moving any VMs is whether every Proxmox node sees the same test LUN through the expected paths. Compare its SCSI WWID across the nodes and confirm that the paths to it resolve to one multipath device per node. Then use a disposable VM to check that I/O continues or recovers as expected when one storage path goes away. If you're carving VM disks out of a shared SAN LUN, Proxmox documents its shared lvm backend for that. An lvmthin pool is local to one node. Proxmox shared storage and multipathing · Proxmox LVM-thin

Are the two 5010s going to serve separate LUNs, or are you planning to use Peer Persistence? A single-path failure test only checks host path redundancy. If the goal is to keep the VM running when an array goes down, I'd confirm the exact PVE kernel, array OS, transport and HPE host toolkit combination with HPE first, then test array switchover with that same disposable VM. HPE documents specific host toolkit requirements for Peer Persistence. Are you connecting the nodes over iSCSI or Fibre Channel? HPE Peer Persistence guide · HPE compatibility matrix

OpenAI Codex assisted with drafting this reply.
 
  • Like
Reactions: TommyL and X25
If @X25 wanted to ask LLM he would have done, stop posting this slop - its disrespectful... Everyone has access to the same LLMs and sometimes more than you... If its not smart enough to include link to proper documentation instead of github, may be find a different one?
 
  • Like
Reactions: X25
One thing I'd check before moving any VMs is whether every Proxmox node sees the same test LUN through the expected paths. Compare its SCSI WWID across the nodes and confirm that the paths to it resolve to one multipath device per node. Then use a disposable VM to check that I/O continues or recovers as expected when one storage path goes away. If you're carving VM disks out of a shared SAN LUN, Proxmox documents its shared lvm backend for that. An lvmthin pool is local to one node. Proxmox shared storage and multipathing · Proxmox LVM-thin

Are the two 5010s going to serve separate LUNs, or are you planning to use Peer Persistence? A single-path failure test only checks host path redundancy. If the goal is to keep the VM running when an array goes down, I'd confirm the exact PVE kernel, array OS, transport and HPE host toolkit combination with HPE first, then test array switchover with that same disposable VM. HPE documents specific host toolkit requirements for Peer Persistence. Are you connecting the nodes over iSCSI or Fibre Channel? HPE Peer Persistence guide · HPE compatibility matrix

OpenAI Codex assisted with drafting this reply
Be using ISCSI, thanks for this, have you run the 5010 in production as was also after some real world feedback as well as theory hits to check out
 
I didnt fully implement Nimble in a pve environment, but in the lab it worked fine as long as you understand the limitations.

1. Without a third party CAF management, you're limited to LVM thick deployment. This makes your hardware snapshots of limited use, and compression/dedup potentially leading to unpredictable performance. search the forum for specific options but the short version is its blockbridge, immature github projects, or unsupported GFS2/OCFS.
2. The Alternative is to map individual LUNs to individual VMs- which is a nightmare to administer at scale. you would still need to build your own guest quiescence orchestration in guest. It possible, but I'd only consider this approach if you have engineering talent and willingness to spend time on a substantial project.
3. PBS will not integrate with your storage the way Veeam does. and Veeam Alletra integration requires VMware AFAIK.
 
We have a pair of 5010 and will be looking at a clustered proxmox setup running a virtual machines

i don't know which model of alletra i have been provided, but it definitely works with debian, with the last release of proxmox and cetrafs is mostly developed on hpe+alletra at the moment.

it rocks.
 
  • Like
Reactions: X25
I didnt fully implement Nimble in a pve environment, but in the lab it worked fine as long as you understand the limitations.

1. Without a third party CAF management, you're limited to LVM thick deployment. This makes your hardware snapshots of limited use, and compression/dedup potentially leading to unpredictable performance. search the forum for specific options but the short version is its blockbridge, immature github projects, or unsupported GFS2/OCFS.
2. The Alternative is to map individual LUNs to individual VMs- which is a nightmare to administer at scale. you would still need to build your own guest quiescence orchestration in guest. It possible, but I'd only consider this approach if you have engineering talent and willingness to spend time on a substantial project.
3. PBS will not integrate with your storage the way Veeam does. and Veeam Alletra integration requires VMware AFAIK.
thanks for these pointers much appreciated
 
i don't know which model of alletra i have been provided, but it definitely works with debian, with the last release of proxmox and cetrafs is mostly developed on hpe+alletra at the moment.

it rocks.
sounds promising, we need to setup a POC for this
 
Interesting .
i'm Running 3 x X3650 M5 's on FC to 2xHPe Alletra 5000, so far only snapshooting windows is problematic (reboot when deleting snapshots)
the 3 nodes are in a clustter, did a reinstall of one node and it went easy back on plase after installing multipath again.

Using Veeam vbr for backup, workers on a VM Nas storage - tests have been fine.
 
Since CetraFS was already mentioned: It should be noted that CetraFS (together with MeshXFS which is a similiar but different project) are two of the "immature github projects" mentioned by @alexskysilk They look quite promising but right now they are still in the early stage of development so I wouldn't trust any production workloads and data with them.

Regarding the lack of snapshot with LVM/thick one possible workaround is to do more backups but depending on your usecase this might not be an viable option. Thanks to "dirty bitmap" incremental backups can be quite fast and together with the live-restore feature of ProxmoxBackupServer you can get your workloads running again (although with limited performance until the restore is completed) in a short time:
https://pve.proxmox.com/wiki/Migrate_to_Proxmox_VE#Alternatives_to_Snapshots

I don't know well this works with other backup software though and depending on your usecase for snapshots it might not even be a viable workaround in the first place.
 
Since CetraFS was already mentioned: It should be noted that CetraFS (together with MeshXFS which is a similiar but different project) are two of the "immature github projects" mentioned by @alexskysilk They look quite promising but right now they are still in the early stage of development so I wouldn't trust any production workloads and data with them.
Indeed.

I'm releasing the next iteration shortly and from a development point of view i'm quicly approaching the maturity that i can get with the resources i have

For anyone who has an alletra or a similar beast, i would really appreciate if you could create a 1T LUN and do experiments with the file system.
Testing and getting feecbacks is the only way forward with this kind of projects.

So while i appreciate those warnings around production, i would also appriate if anyone who backs off also remembers that testing and feedbacks are requested and appreciated.
 
  • Like
Reactions: Johannes S