Ceph RGW
13.7 PBusable capacity
$1.46cost per usable TB‑month, over five years
- Five‑year cost
- $1.2M
- Revenue a year, sold out
- $983K
- Small reads per second
- 10K
- Failed drive rebuilt in
- 2.7 hours
- Metadata flash per server
- 17 TB
- Drives
- 26 TB CMR
Beta Q1 2027 · GA Q2 2027
maxObject
Object storage,
built around the hard drive.
S3 storage for cloud providers. More terabytes to sell, faster service, and full protection back in minutes — on CMR or host‑managed SMR hard drives.1
Baseline
13.7 PB
Ceph RGW
26 TB CMR drives
+35%
18.5 PB
maxObject
26 TB CMR drives
+65%
22.8 PB
maxObject
32 TB host-managed SMR drives
More to sell. Same racks.
On the largest clusters, wider stripes put even more of every drive to work.
CMR and SMR
Every hard drive.
Even the densest.
maxObject runs on conventional CMR hard drives of any size. It also supports host‑managed SMR drives out of the box — they hold more data in every slot, but only accept sequential writes, so most object stores, Ceph RGW included, can’t use them. maxObject streams every write sequentially, so SMR simply works.
- CMR today, SMR when you’re ready. Ceph RGW doesn’t support host‑managed SMR drives. maxObject supports both.
- Mix freely. SMR and conventional CMR drives of any size, in one cluster.
- More in every slot. 1,000 slots of 32 TB SMR drives give 22.8 PB usable, vs 13.7 PB for Ceph RGW on 26 TB CMR drives.2
Hard‑drive support
| Drive type | Ceph RGW | maxObject |
|---|---|---|
| Conventional (CMR) drives | Yes | Yes |
| Host‑managed SMR drives | No | Yes |
| SMR and CMR in one cluster | No | Yes |
Each write would overwrite part of the next track — so a host‑managed SMR drive refuses it.
Writes flow down the zone, track after track — exactly how SMR wants them.
A new kind of object store
Objects, packed together.
Hard drives plus erasure codes give you a lot of space for the money. But many platforms, Ceph RGW included, build an erasure‑code set for every object. Reads must touch every piece, and small objects waste space. maxObject packs objects instead.3
Ceph RGW One erasure‑code set per object
Every object is split across 12 drives, plus 4 for parity. Small objects still take a slice of every drive — the unused part of each slice is wasted.
maxObject Many objects, one wide stripe
Objects are packed whole, side by side, so a read usually touches one drive. Parity protects the entire stripe, and nothing is left empty.
- Your objects
- Other objects
- Parity
- Wasted space
Speed
Faster than Ceph RGW.
By design.
maxObject packs many objects into each erasure‑coded stripe and streams writes. A small read touches one drive instead of twelve, and writes stop keeping the drives busy seeking — so reads get nearly all of their time.4
- Streaming PUTs. Writes are buffered into large regions, protected across nodes in RAM, and streamed to every drive.
- Tiny objects never touch a hard drive. Objects up to 8 KB live with the metadata, on flash.
- Listings and lookups don’t wait on a hard drive. HEAD, LIST and key lookups are answered from flash.
- Multipart uploads complete without copying. The parts are never copied or reassembled.
- Spare RAM and flash become read cache.
Ceph RGW 12+4 per object
12 reads
maxObject packed stripes
1 read
Faster for customers. Easier on disks.
Recovery
Back to full protection
in minutes.
When a drive fails, every healthy drive in the cluster takes a share of the rebuild. The larger the cluster, the faster it finishes — and customers keep reading and writing the whole time.6
- Only live data is rebuilt. A drive that isn’t full rebuilds proportionally faster.
- Customers don’t notice. Data on a failed drive is reconstructed on the fly when it’s read, and new writes carry on.
- A whole server, too. If a server is lost, its metadata is fully protected again within minutes.
- No one has to act. Failed drives are rebuilt, retired and removed automatically.
Minutes
to re‑protect the metadata of a lost server
Service carries on while it heals.
Calculator
Your cluster. Both ways.
Set your servers, drives and price. See what you’d have to sell on Ceph RGW and on maxObject — and how fast a failed drive comes back.
+4.9 PB more to sell
About $350,000 more revenue a year.
And about 32% less cost per usable terabyte.
1,000 drives · estimate7
maxObject
18.5 PBusable capacity
$1.46cost per usable TB‑month, over five years
- Five‑year cost
- $1.1M
- Revenue a year, sold out
- $1.3M
- Small reads per second
- 120K
- Failed drive rebuilt in
- 25 min
- Metadata flash per server
- 1.3 TB
- Drives
- 26 TB CMR
Rebuild of a full drive, cluster otherwise idle · 999 drives share the work
Estimates from published drive characteristics and design arithmetic, not benchmarks. For a sizing of your cluster, talk to our team.
Talk to usFor cloud hosters
Simple to run.
Ready to sell.
One daemon per server. Self‑healing. No placement‑group tuning. Run from maxRAIDOne’s interface and API, and seen in maxRAIDOne Central next to your flash systems. Everything you need to turn disks into a product.
Tenants and billing.
Multi‑tenant accounts with IAM‑style users and policies, quotas, and billing rules you define. Usage is metered every five minutes.
Flash and hard drives, one cluster.
Sell a bulk class on hard drives and a performance class on QLC or PLC flash — one set of accounts and policies.
Clients.
AWS SDKs and CLI work as‑is.
Drives run fuller.
About 85% full, against the roughly 70% Ceph operators plan for — placement is even and recovery needs little spare room.
Leaner servers. Lower cost.
About 1.3 TB of metadata flash per server, against about 17 TB for Ceph RGW — and memory that doesn’t grow with drive count.
One kind of server.
Every server does everything — no gateway, metadata or storage‑only servers to size and buy.
Listing at any scale.
Keys are kept in order, so listing a billion‑object bucket is one ordered walk.
Integrity built in.
Every object, block, metadata record and transfer is checksummed, end to end.
Growth.
Add servers or drives of any size to a live cluster, with no rebalancing storm.
Turn disks into revenue.
maxObject enters beta in Q1 2027, with general availability planned for Q2 2027. Tell us about your object storage and we’ll keep you close.
- maxObject is pre‑release. Features and figures may change before general availability. All comparisons are against Ceph RGW.
- Usable capacity in the same 1,000 drive slots, with protection against any four drive losses in every case. Ceph RGW 12+4 on 26 TB CMR drives; maxObject with 36+4L+3G LRC sets on 26 TB CMR drives, and on 32 TB host‑managed SMR drives.
- Simplified: both drawn 16 drives wide. maxObject’s real stripes are wider (36+4+3).
- One S3 GET of a typical object, as described in the maxObject technical brief.
- Design arithmetic from published drive characteristics and IBM Cloud Object Storage’s published traces, not benchmarks: 1,000 drive slots (50 servers × 20 drives), Ceph RGW at 12+4. Weighted by a week of real public‑cloud traffic, maxObject needs about one sixth of the drive time.
- Design arithmetic, not a benchmark: a full 26 TB drive in a cluster of 50 servers with 24 drives each, at 12+4, limited by drive speed (about 27 minutes); about 18 minutes is the network limit at 2 × 25 GbE per server. Smaller clusters take longer; drives that aren’t full take less time. Under heavy customer load the rebuild yields and takes longer.
- Estimate, not a benchmark. Usable capacity = raw capacity × the share of each stripe that holds data × how full drives run: Ceph RGW at 12+4 and 70% full; maxObject with the protection you choose and 85% full. Every option survives any four failures, with one stripe member per server. Revenue assumes all usable capacity is sold. Small reads: about 120 per drive per second for maxObject and 10 for Ceph RGW. Metadata flash scales from 1.3 TB vs 17 TB per server of 20 × 26 TB drives; 17 TB is the middle of the 13–21 TB that Ceph’s published guidance implies. Rebuild: a full drive, shared by every other drive, on an otherwise idle cluster, or limited by the network; it takes longer under heavy customer load. maxObject streams at about 175 MB/s per drive. Ceph RGW at 12+4 reads each 4 MiB RADOS object as pieces of about 340 KiB, each costing an 8 ms seek: about 35 MB/s of useful repair per drive. Ceph’s default recovery limits often make it slower in practice. Five‑year cost: the same servers and hard drives for both, priced as you set them (SMR drives at the same price per TB as CMR); metadata flash as above; power at 8 W per hard drive and 250 W per server, plus 40% for cooling. Excludes software, rack space and staff. Cost per usable TB‑month = five‑year cost ÷ usable capacity ÷ 60 months.