Hi community!
I'm new to Proxmox VE and would like to get some validation of a ZFS cluster configuration.
My goal is to host 50 production VMs (Linux and Windows servers).
Initially, I considered a 3-node hyperconverged Ceph but changed my mind after realizing that a cluster this small would become a headache to manage if (when) a node fails or is shutdown for maintenance.
Hence, to keep it simple and reliable, I decided to go with 2 identical nodes with local ZFS, replicating VMs from one to the other. I'm OK with a 15-minute RPO in case of a node failure (cluster HA).
Here is the hardware spec I came out with for each node:
Dell R650
02 Xeon Gold 5317 (12 Cores - 3 GHz)
384 GB RAM (average RAM usage per VM - 4 GB)
02 240 GB M.2 NVMe Sticks (boot pool - mirror)
08 1,92TB SSD SATA Read Intensive (storage pool - 4 mirror pairs)
01 400 GB NVMe Write Intensive with Power Protection (SLOG - lots of sync writes)
01 HBA 330 Controller (for the SSD SATA disks)
04 10GbE ports (2 for management network, 2 for uplinks)
Questions:
1. Is this a sane specification for the ZFS cluster?
2. Would it be OK to replace the 08 SSD SATA disks with faster NVMe disks and use ZIL on zpool instead of SLOG?
Thanks in advance!
I'm new to Proxmox VE and would like to get some validation of a ZFS cluster configuration.
My goal is to host 50 production VMs (Linux and Windows servers).
Initially, I considered a 3-node hyperconverged Ceph but changed my mind after realizing that a cluster this small would become a headache to manage if (when) a node fails or is shutdown for maintenance.
Hence, to keep it simple and reliable, I decided to go with 2 identical nodes with local ZFS, replicating VMs from one to the other. I'm OK with a 15-minute RPO in case of a node failure (cluster HA).
Here is the hardware spec I came out with for each node:
Dell R650
02 Xeon Gold 5317 (12 Cores - 3 GHz)
384 GB RAM (average RAM usage per VM - 4 GB)
02 240 GB M.2 NVMe Sticks (boot pool - mirror)
08 1,92TB SSD SATA Read Intensive (storage pool - 4 mirror pairs)
01 400 GB NVMe Write Intensive with Power Protection (SLOG - lots of sync writes)
01 HBA 330 Controller (for the SSD SATA disks)
04 10GbE ports (2 for management network, 2 for uplinks)
Questions:
1. Is this a sane specification for the ZFS cluster?
2. Would it be OK to replace the 08 SSD SATA disks with faster NVMe disks and use ZIL on zpool instead of SLOG?
Thanks in advance!