Against Ceph with three copies
maxRAID Grid vs Ceph
Resilient block storage across many servers, for cloud and AI.
wildfire-storage.com/grid
one 8+2 arraythe next arrayMmaster serverspare server
Features and figures may change. maxRAID Grid performance figures are design targets.
At a glance
More flash. More speed. Lower cost.
Cost per usable TB, five years, all in
DrivesParity drivesLicenceMaintenance
Shown with 2 parity drives, as resilient as Ceph’s 3 copies. Your choice: 1, 2, 3 or more.
| Ceph | maxRAID Grid | |
|---|---|---|
| Protection | 3 full copies (33%) | 8 data + 2 parity (80%) |
| How full it runs | 70% | 90% |
| Usable share of raw flash | 23.3% | 72.0% |
| Cost per raw TB | $500 | $500 SSD + $120 licence |
| Maintenance | — | $18 per raw TB a year |
$986
per usable TB over five years. Ceph: $2,143.
3.1×
the usable capacity, same flash
26×
the writes per server, and 5× the reads
SSDs at $500 per raw terabyte. Drives and parity drives at that price (in this example, 2 of every 10 drives are parity; you choose how many). maxRAID Grid list prices: licence $120 per raw terabyte, parity drives included, plus maintenance $18 per raw terabyte a year, five years counted. Ceph is counted with no licence or support cost. Usable terabytes = raw × share that holds data × how full it runs. maxRAID Grid speed figures are design targets.
2 / 6Performance per server
5× the reads. 26× the writes.
Handle the same load with fewer servers.
Small blocks (4 KB): operations per second
Reads
Writes
Large blocks (64 KB): GB per second
Reads
Writes
maxRAID GridCeph, 3 copies
Ceph (measured): “Ceph: A Journey to 1 TiB/s”, ceph.io, January 2024: 63 servers, each with 10 NVMe SSDs and 2 × 100 GbE, 3 copies, tested on the storage servers themselves; per server = cluster result ÷ 63. The Ceph 64 KB bars are estimates: “Ceph Reef Freeze Part 1” (ceph.io, 2023) measured 128 KB random writes at about 2.3 GB/s per server; reads taken as about 3× writes. maxRAID Grid: per‑server design targets, with the same 10 NVMe SSDs and 200 GbE of networking per server.
3 / 6Resilience
Lose a drive. Lose a server. Keep serving.
One 8+2 arrayServer 4 downSpare server
Losing a server costs each array just one drive.
With 2 parity drives, any two drives or servers can fail. Or pick 1, 3 or more.
Drives rebuild from the other servers. A spare server steps in.
10 servers × 10 SSDs, with one spare server. Each logical array (8 data + 2 parity drives) has one drive in every server; a failed drive is rebuilt onto spare capacity from its array’s drives in the other servers, while the array keeps serving. The number of parity drives is your choice; 2 is shown because it matches the resilience of Ceph’s three copies. Features and figures may change.
4 / 6How it fits
Fits what you run. Grows with you.
Standard x86, any SSD, Ethernet.
Volumes on demand, through a CSI driver.
NVMe over TCP or RDMA, or iSCSI. Shared volumes for AI.
Data rebalances automatically.
The Kubernetes driver ships with general availability. RDMA = direct memory access across the network; iSCSI = block storage over standard IP networks. Features and figures may change.
5 / 6Next step
See it run. On your servers.
How an evaluation works
- 1We size it with you.
- 2We set it up on your servers.
- 3You test it your way.
Talk to us any time.
wildfire-storage.com/grid
sales@wildfire-storage.com
+1‑707‑maxRAID (707) 629‑7243
Features and figures may change.
6 / 6