1.2 Petabytes Per Hour: Backup Has Never Been This Fast

StorageReview
StorageReviewApr 3, 2026

Why It Matters

The HPE accelerator dramatically shortens backup windows, enabling enterprises to protect AI‑generated data without sacrificing SLAs, while forcing a broader shift toward high‑speed networking and application tuning.

Key Takeaways

  • HPE's Data Protection Accelerator delivers 1.2 PB/hour backup speed.
  • Accelerator node offloads deduplication, boosting X1000 throughput significantly.
  • Four DPEs backed 1,440 VMs in ~40 minutes.
  • Linear scaling achieved with up to ten accelerator nodes.
  • Network fabric upgrades required to fully exploit performance gains.

Summary

HPE showcased a breakthrough in backup and recovery by pairing its all‑flash X1000 storage platform with a new Data Protection Accelerator (DPE) node. In a Raleigh lab the team wired three to four racks of gear—30 ESXi hosts, roughly 1,440 virtual machines, and multiple X1000 clusters—to stress‑test the solution, aiming to prove that backup can keep pace with modern data growth. The test demonstrated headline‑grabbing numbers: the accelerator cluster achieved 1.2 petabytes per hour, equivalent to about 300 TB per hour per DPE, and completed a full backup of the 1,440 VMs in roughly 38‑40 minutes. By offloading deduplication, encryption, and tagging to the DPE, the X1000 storage ceased to be the bottleneck, allowing linear performance scaling as additional accelerator nodes are added. Engineers highlighted that the DPE acts as a compute‑heavy front‑end, reducing the data volume that reaches the primary array. “Speed is the end‑to‑end story,” one speaker noted, emphasizing that the architecture not only accelerates ingest but also shortens recovery times—provided the network fabric can sustain multiple 25 Gbps links and downstream servers can read the data fast enough. The lab also uncovered operational challenges, such as the need to upgrade Ethernet fabrics and tune backup applications to exploit the newfound throughput. For enterprises wrestling with exploding data volumes—driven by AI model training, analytics, and compliance—this architecture promises to shrink backup windows dramatically, preserve service‑level agreements, and potentially repurpose the high‑performance storage for other workloads. However, realizing the full benefit will require coordinated upgrades across storage, networking, and application layers.

Original Description

We traveled to HPE's lab in Raleigh, North Carolina to stress test the Data Protection Accelerator Node (DPAN) paired with the Alletra Storage MP X10000, a flash-native, scale-out object storage platform. The goal was simple: find out how fast modern backup can actually go when the storage target is no longer the bottleneck.
The test environment included 30 ESXi hosts, 1,440 VMs, four DPANs, and an 8-node X10000 cluster across roughly three and a half racks of gear. HPE didn't just talk numbers on paper, they built the environment and let us push it.
In our focused three-server validation, a single DPAN completed a full backup of 144 VMs (27.36 TB) in 26 minutes at 17.54 GB/s. Scaling to seven servers in the "hero configuration," a single DPAN hit 83.83 GB/s (301.7 TB/hr) across 336 VMs in just 13 minutes, and still hadn't reached its ceiling. The constraint shifted to backup application orchestration, not storage. A separate Oracle RMAN test showed a 100 TB database backup completing in 2.4 hours with DPAN vs. 4.3 hours direct-to-X10000, a 1.79x throughput improvement.
Three key infrastructure takeaways emerged: backup applications aren't tuned for this level of throughput, network fabrics need to match (multiple bonded 25 GbE links, not legacy 10 GbE), and recovery speeds can outpace client-side write performance on the servers being restored.
Read the full paper:
Follow StorageReview:
#DataProtection #HPEStorage #EnterpriseBackup
0:00 Introduction from HPE Raleigh
0:56 Lab walkthrough with Chad
1:19 Why DPAN matters: speed
2:06 How the accelerator node works
3:29 Test environment: 30 hosts, 1,440 VMs
4:08 Taking performance off paper
5:02 1.2 PB/hr and why HPE quotes numbers
5:43 AI, data growth, and backup windows
7:16 Availability and customer adoption
7:51 Kevin's hardware deep-dive with Bob Jensen
10:30 First full backup results
11:08 Back in the StorageReview lab
13:29 Three infrastructure readiness takeaways
15:31 Recovery challenges and scaling with DPAN

Comments

Want to join the conversation?

Loading comments...