ฉันจะโฮสต์ S3-compatible object storage ของตัวเองบนเซิร์ฟเวอร์เช่าได้อย่างไร?

ฉันควรใช้ S3 server ตัวไหนบนเครื่องเดียว?

เลือก S3 server ตามคุณสมบัติที่ไคลเอนต์ของคุณต้องการ แล้วกำหนดขนาดเครื่อง เวอร์ชันด้านล่างคือ stable release ล่าสุดบนหน้าของโปรเจกต์เอง อ่านเมื่อ 2 พฤศจิกายน 2026

Erasure coding หรือ ZFS: redundancy ควรอยู่ที่ชั้นไหน?

Put redundancy in one layer only. Erasure coding on top of RAIDZ2 pays for parity twice and gains little on a single machine.

Design A: separate disks, erasure coding in the object store. MinIO, SeaweedFS and Ceph each split objects into data and parity shards across the drives. A failed drive costs only the shards on it, and the store heals object by object.

Design B: a ZFS pool, no redundancy in the object store. Build RAIDZ2, then run Garage with one replica or SeaweedFS without replication. ZFS checksums every block, repairs from parity, and gives you snapshots and compression on the whole pool.

Our view: on one rented machine, Design B is simpler to run and easier to rescue. One zpool status shows the health of every drive. ZFS snapshots also guard against a deleted bucket. Design A wins when you want per-object healing or plan to add machines later.

Garage needs Design B. Its quick start says a single-node layout gives no redundancy at all. With one copy, Garage's checksums can detect a bad block but cannot rebuild it.

แต่ละเลย์เอาต์เหลือพื้นที่ใช้งานได้เท่าไร?

Usable space depends on how many drives hold parity. OpenZFS gives the RAIDZ rule: N disks with P parity hold about (N − P) × the disk size. MinIO's default parity is EC:4 for sets of 8 drives or more and EC:2 for 5 or fewer. SeaweedFS's 10+4 code stores about 1.4 times the data.

How long does the first upload take?

At 1 Gbit/s, the best case is 125 MB per second. Filling Hold 100 after RAIDZ2 (80 TB) takes at least 7.4 days. Filling Hold 160 (128 TB) takes at least 11.9 days.

These are line-rate calculations, not measurements: real transfers run slower. Each hold's monthly allowance is larger than one full fill: ทุกเซิร์ฟเวอร์มีพอร์ต 1 Gbit/s พร้อมโควตาทราฟฟิกรายเดือน (Hold 24 40 TB, Hold 40 60 TB, Hold 100 120 TB, Hold 160 200 TB), IPv4 address แบบเฉพาะหนึ่ง address และ IPv6 /64; การใช้เกินโควตาจะไม่ทำให้เกิดการเรียกเก็บเงินอัตโนมัติ. Plan a big migration in parallel streams, for example with rclone copy --transfers 16.

How do I set up Garage on a ZFS pool?

This is Design B on Hold 100: ten data drives in RAIDZ2, metadata on the mirrored system SSDs. Garage's deployment guide puts metadata on SSD and data on hard drives. Replace the device names and secrets with your own.

# 1. RAIDZ2 pool on the ten blank data drives (use /dev/disk/by-id names)
zpool create -o ashift=12 tank raidz2 /dev/disk/by-id/ata-HDD1 /dev/disk/by-id/ata-HDD2 ... /dev/disk/by-id/ata-HDD10
zfs create -o compression=lz4 -o atime=off tank/garage

# 2. /etc/garage.toml (metadata on the system SSD, data on the pool)
metadata_dir = "/var/lib/garage/meta"
data_dir = "/tank/garage"
db_engine = "sqlite"
replication_factor = 1
rpc_bind_addr = "[::]:3901"
rpc_public_addr = "127.0.0.1:3901"
rpc_secret = "<openssl rand -hex 32>"

[s3_api]
s3_region = "garage"
api_bind_addr = "[::]:3900"

[admin]
api_bind_addr = "127.0.0.1:3903"
admin_token = "<openssl rand -base64 32>"

# 3. Layout, bucket and key
garage status
garage layout assign -z dc1 -c 70T NODE_ID   # NODE_ID from garage status; leave headroom
garage layout apply --version 1
garage bucket create backups
garage key create backup-key
garage bucket allow --read --write --owner backups --key backup-key

Is a rented server cheaper than cloud object storage?

Per terabyte, the ten-drive holds cost less than Backblaze B2 or Wasabi, and the server bills no egress or API calls. Hold 100 costs $3.77/TB per raw TB and Hold 160 $3.50/TB. As of 2 November 2026, Backblaze B2 lists $6.95 per TB a month and Wasabi $7.99.

Raw is not usable, though. RAIDZ2 on ten drives keeps 80%, so each usable TB costs a quarter more than each raw TB. Both ten-drive holds stay below B2 even then. On a two-drive hold a mirror halves the space, and the cloud wins. The cloud also keeps your data in more than one place, which one server does not. The full sum, with egress, is on is a rented storage server cheaper than Backblaze B2 or Wasabi. Every hold is on the Basalthold pricing page.