From pg at glfs.list.sabi.co.UK Fri Dec 12 06:04:46 2025 From: pg at glfs.list.sabi.co.UK (Peter Grandi) Date: Fri, 12 Dec 2025 06:04:46 +0000 Subject: [Gluster-users] using devices with dfifferent sizes Message-ID: For the first time I am considering setting up a volume with devices of different size that is 40 4TB SSDs plus 24 8TB SSDs and I wonder how easy it is to keep them balanced, My intended setup is 3-way replication with sharding. Will the sharding. Possible configurations: * All 64 SSDs belong to one volume. * The 8TB SSDs are split into 2 partitions of 4TB each. and all 88 4TB partitions belong to one volume. * Two volumes one of 40 4TB SSDs one of 24 8TB SSDs. Please let me know pluses/minuses/recommentations. Notes: These are SSDs so 2 partitions on one is fine. If they were HDDs I would not do that. SSD sizes are rounded. From pg at glfs.list.sabi.co.UK Mon Dec 8 23:52:40 2025 From: pg at glfs.list.sabi.co.UK (Peter Grandi) Date: Mon, 08 Dec 2025 23:52:40 +0000 Subject: [Gluster-users] filesystem type: Btrfs, ZFS Message-ID: So I am going to setup one or more volumes on 40-60 4TB-8TB SSDs (unfortunately without PLP but high speed is not a requirement). For "reasons" I would use a filesystem type with built-in compression and in practice that means Btrfs or ZFS: cacheFS is a bit immature still and compression in F2FS (which is otherwise a very good choice in most cases whether for HDD or SSD) is a bit weird. I have used bothb extensively and I would use both on a single SSD, so no use of their built-in volume manager. The question is which one works better with GlusterFS and I guess a significant aspect is how big the GlusterFS extended attributes are "typically" and whether that much can be inlined in an i-node. ZFS inodes can be pretty big (16KiB) and inline a lot of data; but inline extended attributes for Btrfs IIRC are limited to around 3900B on systems with 4KiB pages. Any other considerations?