6.2 HPE Alletra Storage MP Hardware: Controller Nodes, JBOFs, Switchless & Switched Systems
Key Takeaways
The Alletra Storage MP building block is a 2U chassis that holds either two controller nodes or 24 NVMe SSDs; controllers and JBOF shelves connect over 100 GbE.
B10000 switchless systems (B10120, B10130, B10140) put 2 nodes and 24 drive bays in one 2U chassis; a 4-node B10140 (4N) uses two chassis.
B10000 switched systems (B10230, B10240) use a pair of dedicated 100GbE switches, 2 to 6 controller nodes and up to 16 JBOFs, for about 5.9 PB raw.
B10000 controller nodes have no built-in host ports; each needs at least one host adapter for FC/NVMe-FC or Ethernet iSCSI/NVMe-TCP.
The X10000 scales from 3 to 16 controller nodes (64-core, 512 GB each) and 1 to 16 fully populated 24-drive JBOFs, about 23.6 PB raw.
HPE Alletra Storage MP Hardware
Quick Answer: Alletra Storage MP is built from 2U chassis that hold either two controller nodes or 24 NVMe SSDs (a JBOF). In a switchless B10000 the controllers and up to 24 drives share one chassis and JBOFs connect directly. In a switched B10000, two dedicated 100GbE switches connect 2 to 6 controller nodes to up to 16 JBOFs, so performance and capacity scale independently. The X10000 uses the same building blocks with 3 to 16 nodes.
Why disaggregation matters
In a traditional dual-controller array, controllers and drives are bought together. If you need more performance you buy a bigger array; if you need more capacity you may also buy controller power you do not need. In Alletra Storage MP the compute (controller nodes) and the capacity (NVMe JBOFs) are separate units on a shared network. In a switched configuration every controller node can reach every JBOF, so adding either one rebalances the system without forklift upgrades.
B10000 controller node models
| Model | Configuration | Nodes | Cores per node | Cache memory per node | Drive shelves |
|---|---|---|---|---|---|
| B10120 | Switchless | 2 | 8 | 256 GB | Base chassis only |
| B10130 | Switchless | 2 | 16 | 256 GB | Base plus up to 2 JBOFs |
| B10140 | Switchless | 2 | 32 | 512 GB | Base plus up to 2 JBOFs |
| B10140 (4N) | Switchless, two 2U chassis | 4 | 32 | 512 GB | 2 base chassis plus 1 JBOF (2 JBOFs, up to 96 drives, with ArcusOS 10.6.0) |
| B10230 | Switched | 2 to 6 | 16 | 256 GB | Up to 16 JBOFs |
| B10240 | Switched | 2 to 6 | 32 | 512 GB | Up to 16 JBOFs |
Each node has one CPU (HPE uses AMD EPYC embedded processors) and four slots. All nodes in a system must be the same type. ArcusOS 10.6.0 added five- and six-node switched systems and a non-disruptive conversion from switchless to switched, so a customer can start small and grow.
Capacity
| Configuration | Maximum raw capacity | Maximum effective capacity |
|---|---|---|
| Per shelf (24 drives) | 737.2 TB | 1,993.5 TB |
| B10130 or B10140 switchless system | 2,211.8 TB | 6,165.6 TB |
| B10230 or B10240 switched system | 5.9 PB | 16.6 PB |
HPE calculates effective capacity with a 4:1 data compaction ratio (thin provisioning, deduplication, compression and copy technologies) in RAID 6 (10+2), with 9% free-space headroom. Actual savings depend on the workload.
Chassis, slots and adapters
- Switchless chassis: two controller nodes plus 24 SFF NVMe drive bays.
- Switched chassis: two controller nodes and no drives; all SSDs live in JBOFs.
- Slots: each node has four OCP slots. Slot 2 (and optionally slot 1) holds the 100GbE back-end adapter, and slots 3 and 4 (and slot 1 when it is not used for the back end) hold host adapters. The exact slot rules depend on the model and ArcusOS release, so check the QuickSpecs.
- No built-in host ports: every node needs at least one host adapter.
| Host adapter type | Protocols |
|---|---|
| 32/64Gb Fibre Channel | FC and NVMe/FC; current systems support up to 12 FC ports per node |
| 10/25GbE Ethernet | iSCSI and NVMe/TCP |
| 100GbE Ethernet | iSCSI and NVMe/TCP |
| 10/25GbE replication adapter | Optional Remote Copy over IP (RCIP) |
For unified file, the Ethernet adapter must be in slot 4, with ports 1 and 2 dedicated to file.
JBOF expansion shelves
- 2U, 24 NVMe SFF bays, two shelf nodes with 8-core CPUs and 64 GB of memory that is not used as cache.
- Each shelf node connects with a 100GbE OCP adapter.
- Drives are added from 8 to 24 per enclosure in two-drive increments, and JBOFs are added one at a time.
Switchless versus switched B10000
| Attribute | Switchless (B10100 family) | Switched (B10200 family) |
|---|---|---|
| Back-end network | Direct 100GbE cables between nodes and JBOFs | Two dedicated 32-port 100GbE switches |
| Drives in the controller chassis | Yes, up to 24 | No |
| Controller nodes | 2, or 4 with the B10140 (4N) | 2 to 6 |
| JBOFs | Up to 2 (on the 4-node model, 1, or 2 with ArcusOS 10.6.0) | Up to 16 |
| Best for | Smaller systems, lower cost | Large scale and independent growth |
The switches are dedicated to back-end traffic between controller nodes and JBOF shelf nodes. HPE states they cannot be used to connect hosts or any other systems; hosts connect through the host adapters to the customer's own SAN or Ethernet network.
Drive rules
| Rule (B10000 QuickSpecs) | Value |
|---|---|
| Drive interface | NVMe |
| NAND types | TLC or QLC; no mixing in one system |
| Minimum | 8 drives per chassis or system |
| Increments | 2 drives |
| Capacities | All drives in a system must be the same capacity and type |
| Encrypted options | Self-encrypting and FIPS-validated SSDs; encryption needs a license |
| Protection | RAID 6 only |
QLC SSDs give lower cost per TB and higher density for read-heavy workloads; TLC SSDs suit write-intensive work.
X10000 hardware
| Component | X10000 value (QuickSpecs) |
|---|---|
| Controller chassis | 2U, one or two controller nodes; 2 to 8 chassis |
| Controller nodes | 3 to 16; one 64-core CPU and 512 GB RAM each |
| Back-end | Two 2-port 100GbE adapters per node; a pair of internal fabric switches (100 GbE, with a 400 GbE bundle available) |
| Front-end | 100GbE or 10/25GbE host ports (no mixing within a cluster) |
| JBOFs | 1 to 16; 24 NVMe drives each, fully populated |
| Drives | TLC 3.84 to 30.72 TB, or QLC 15.36 to 61.44 TB; no TLC and QLC mixing in a cluster |
| Raw capacity | About 23.6 PB per cluster (384 drives) |
| Effective capacity | Calculated at 2:1 compaction with triple-parity (20+3+1) protection |
| Optional nodes | Data intelligence nodes (ProLiant servers with NVIDIA GPUs) and data protection accelerator nodes |
The smallest X10000 has three controller nodes, so it needs two controller chassis.
Worked sizing example
A customer starts with a B10140 switchless system with 12 x 7.68 TB TLC SSDs (92 TB raw). Two years later it needs more capacity: it adds drives in pairs up to 24 in the base chassis, then adds a JBOF. When it later needs more controller performance and more than two JBOFs, ArcusOS 10.6.0 allows a non-disruptive conversion to a switched configuration.
Common exam traps
- Hosts on the back-end switches: not allowed; they are internal only.
- Drives in a switched controller chassis: none; use JBOFs.
- Mixing TLC and QLC or capacities: not supported in a B10000 system.
- Built-in host ports: there are none; order host adapters.
In a switched HPE Alletra Storage MP B10000, where are the NVMe SSDs installed?
In the controller node chassis, up to 24 drives in each chassis
Only in the JBOF expansion shelves; the controller chassis holds no drives
In the 100GbE back-end switches that connect the nodes
In an attached MSA 2U SFF expansion enclosure
A customer wants to connect application servers directly to the two 32-port 100GbE switches supplied with a switched B10000 to save buying SAN switches. What should the architect advise?
Connect them; the switches support up to 16 directly attached servers
Connect them, but only for hosts that use NVMe/TCP
Connect hosts to one switch and keep the other for back-end traffic
Advise against it: the switches are for back-end traffic only
A B10000 system has twelve 7.68 TB TLC SSDs. The customer wants to add three 15.36 TB QLC SSDs. Which rules does this violate?
Two-drive increments, one capacity and type, and no TLC/QLC mixing
None; the B10000 supports any mix of drive types and capacities
Only the rule that QLC drives must sit in an MSA expansion enclosure
Only the requirement to protect QLC drives with RAID 10 mirroring
What is the minimum number of controller nodes in an HPE Alletra Storage MP X10000 cluster?
1 node
2 nodes
3 nodes
16 nodes
Sections you finish are checked off in the contents.