High Performance Storage System

Incremental Scalability
Based on storage needs and deployment schedules, HPSS scales incrementally by adding computer, network and storage resources. A single HPSS namespace can scale from petabytes of data to exabytes of data, from millions of files to billions of files, and from a few file-creates per second to thousands of file-creates per second.
About HPSS   :    The Big Picture


Below is the big picutre of an HPSS storage system, with overview details of the user interfaces supported by HPSS (at the top); an illustration the showcases our use of IBM Db2 for metadata on flash storage; the RHEL platforms and silicon technologies supported by HPSS; the incrementally scalable disk and tape hierarchies that point to the disk and tape storage hardware. For additaionl details, select one of the tabs.

View the big picture in a separate window

Red Hat Enterprise Linux (RHEL) Platforms

Hardware Vendor Neutral

The HPSS Collaboration agrees that the HPSS software shall remain hardware vendor neutral. HPSS Core Server and HPSS Mover software runs on Red Hat Enterprise Linux platform, and supports x86-64 and OpenPower servers. Customers run HPSS software on Dell, HP, IBM, Lenovo, and other popular enterprise class rack mounted servers.

Core Server
RHEL 6 & 7
AMD x86-64
Intel x86-64
IBM OpenPower
HPSS 7.5 RHEL platform compatibility matrix

Metadata Storage

HPSS File Operations Per Second

Every HPSS file operation requires one or more Db2 database transactions (e.g. an HPSS file create with data requires four Db2 metadata transactions). HPSS services help with planning, provisioning, and configuring Db2 to meet your transaction requirements.

HPSS Backed by Db2

Metadata describing the files in HPSS are managed by Db2 relational databases. Each relational database has tables and indexes, transaction log files, archived transaction log files , and database backup files. All Db2 transaction log files and the Db2 database backup files are often called the Db2 recovery objects.

Mandatory HPSS Redundancy

When Db2 metadata must survive a complete chassis failure, a second local server is provisioned with the RAID-device for the second copy of Db2 recovery objects discussed earlier. The Db2 transaction log files are further protected by configuring Db2 log mirroring to make a second copy on a separate RAID-device. The Db2 log archive files are also mirrored to a separate RAID-device. All Db2 recovery files must be protected using your existing site disaster recovery plan and calls for the HPSS Db2 files to be stored on remote-disk or remote-tape. Your site's bac kup serves as the second copy of the Db2 database backup files. It is not acceptable to store the Db2 backups into HPSS.

Optional HPSS Metadata Redundancy

When Db2 metadata must survive a complete chassis failure, a second local server is provisioned with the RAID-device for the second copy of Db2 recovery objects discussed earlier. The separate chassis can be mounted in a separate rack to protect the HPSS metadata from a rack failure.

Db2 log shipping is used to protect HPSS metadata from a wider range of failures, including a site failure, and to help reduce downtime caused by these failures. Db2 log shipping is used to replicate HPSS Db2 metadata spanning geography.

HPSS Db2 Capacity

Here is an example of Db2 storage for 1 billion HPSS files in a single chassis metadata configuration:
  1. High IOPS storage (at least two RAID devices)
    1. 1 TB RAID: Db2 transaction log files and half of the Db2 tables and indexes
    2. 1 TB RAID: Db2 transaction log mirror files and half of the Db2 tables and indexes
  2. High-bandwidth storage (at least two RAID devices)
    1. 4 TB RAID: Db2 archived transaction log files and half of the Db2 database backup files
    2. 4 TB RAID: Db2 archived transaction log mirror files and half of the Db2 database backup files
  3. Integrating Db2 recovery files into site backup
    1. Db2 recovery files are typically sent to tape using backup software (e.g. Bacula)
    2. 8 TB stores about seven days of Db2 recovery files
SSD vs. HDD for metadata storage

The solid-state disks (SSD) are cost-effective for the high IOPS (Db2 tables, indexes and logs), high bandwidth (for faster backup/restore times), and low bit-error-rate (high integ rity) requirements of HPSS Db2. The new NVMe (non-volatile memory express) is a host controller interface and storage protocol created to accelerate the transfer of data to-and-from solid-state drives (SSD s) over a computer's high-speed Peripheral Component Interconnect Express (PCIe) bus and can be found on modern RHEL Platforms. SSDs deliver tens of thousands of write IOPS, hundreds of thousands of read I OPS, and transfer rates approaching a gigabyte per second per device. Spinning hard disk drives (HDD) are an ideal solution for high capacity data that are not frequently accessed, and are no longer ideal for HPSS Db2 metadata.

Disk and Tape Storage Hardware

HPSS Disk Cache: HPSS supports raw block storage devices like DDN SFA-12K, NetApp E5600, IBM FlashSystem V7000, IBM V5030 and similar enterprise class block storage units. Additionally, HPSS supports block device files on file systems such as OpenZFS and Spectrum Scale.

HPSS Tape Libraries: HPSS software is tested with IBM, Oracle, Quantum and Spectra Logic tape libraries. Contact us if support for another library vendor is desired.

HPSS Tape Drives: HPSS software is tested with LTO and Enterprise tape drives from the tape library vendors we support.

User Interfaces

HPSS Storage Broker

HPSS Storage Broker is used to store, protect, and error correct project datasets across a wide variety of archive storage including public and private S3 object stores, file systems and HPSS. The limited availability release of HPSS Storage Broker was provided to HPSS customers upon request in 4Q 2020, and will be made generally available in 3Q 2021.


SwiftOnHPSS for OpenStack Swift is an S3 interface for HPSS that supports automatic class of service (COS) selection, automatic HPSS end-to-end data integrity support with OpenStack Swift md5 object checks ums, and shared access of Swift objects by other HPSS interfaces.

Spectrum Scale

Intended for HPC use, HPSS can be coupled with Spectrum Scale (previously named GPFS) to automatically: copy files from Spectrum Scale to HPSS; purge Spectrum Scale files that are not being used when space thresholds are reached; recall files from HPSS when accessed by Spectrum Scale users; and save a point-in-time snapshot of Spectrum Scale. HPSS for Spectrum Scale allows multiple Spectrum Scale file syste ms to be managed by a single HPSS.


Linux applications benefit from a near-POSIX standard read-write file system interface. This interface enables HPSS to be mounted as a Linux file system in user space (FUSE). Customers are using HPSS FUSE with Open SSL (encrypted file transfer solution), MinIO (S3 object storage solution), OpenStack (object storage solution), SaMBa (MS Windows file sharing), NFS (POSIX file sharing), DSpace (restful open di gital repository solution), and Bacula (site backup solutions).


The high performance Parallel FTP (PFTP) interface moves files in and out of HPSS at high data rates. Standard FTP and high-performance parallel FTP commands are both supported.


The Hierarchical Storage Interface (HSI) provides a familiar UNIX shell-style interface for managing and transferring files. HPSS parallel file transfers are done automatically. HTAR is a utility for storing groups of files using the POSIX TAR specification and a high performance multithreaded buffering scheme to transfer files directly to and from HPSS. Online documentation for HSI and HTAR is found here.


The Client API is the most powerful interface in terms of control, performance, and rich functionality. The HPSS Client API is the foundation of every HPSS interface, and customers have ported open source applications including:
HPSS Client Interface Compatibility Matrix

Ubuntu 18.04
RHEL 7 & 8
HPSS 9.1 client interface compatibility matrix
  1. FUSE servers available on Red Hat Enterprise Linux 64-bit kernels.
  2. HSI gateway available on Red Hat Enterprise Linux 64-bit kernels. HSI and HTAR clients run on a number of platforms
  3. HPSS User Interface Client support for operating systems not listed in the table above may be provided by special services. See HPSS Offerings for offering details, and contact us.
  4. The PIO API requires the Client API.

< Home

Come meet with us!
2022 HUF
The 2022 HPSS User Forum (HUF) will be an in-person event scheduled October 24-28, 2022, in Houston, TX. Please check back for registration details. This will be a great opportunity to hear from HPSS users, collaboration developers, testers, support folks and leadership (from IBM and DOE Labs). Please contact us if you are not a customer but would like to attend.

The 2022 international conference for high performance computing, networking, storage and analysis will be in Dallas, TX from November 14th through 17th, 2022 - Learn More. As we have each year (pre-pandemic), we are scheduling and meeting with customers via IBM Single Client Briefings. Please contact your local IBM client executive or contact us to schedule a HPSS Single Client Briefing to meet with the IBM business and technical leaders of HPSS.

HPSS @ STS 2023
The 4th Annual Storage Technology Showcase has been postponed, but HPSS expects to support the event when it returns. Check out their web site - Learn More.

HPSS @ MSST 2023
The 37th International Conference on Massive Storage Systems and Technology will be in Santa Clara, California in May of 2023 - Learn More. Please contact us if you would like to meet with the IBM business and technical leaders of HPSS at Santa Clara University.

HPSS @ ISC 2023
ISC 2023 is the event for high performance computing, machine learning, and data analytics, and will be in Hamburg, Germany from May 21st through May 25th, 2023 - Learn More. As we have done each year (pre-pandemic), we are scheduling and meeting with folks attending the conference. Please contact us meet with the IBM business and technical leaders of HPSS.

What's New?
Celebrating 30 Years - 2022 marks the 30th anniversary of the High Performance Storage System (HPSS) Collaboration.

Lots of Data - In March 2022, IBM/HPSS delivered a storage solution to a customer in Canada, and demonstrated a sustained tape ingest rate of 33 GB/sec (2.86 PB/day peak tape ingest x 2 for dual copy), while simultaneously demonstrating a sustained tape recall rate of 24 GB/sec (2.0 PB/day peak tape recall). HPSS pushed six 18-frame IBM TS4500 tape libraries (scheduled to house over 1.6 Exabytes of tape media) to over 3,000 mounts/hour.

HPSS 9.3 Release - HPSS 9.3 was released on December 14th, 2021 and introduces eight new features and numerous minor updates.

HUF 2021 - The HPSS User Forum was hosted virtually at no cost in October 2021.

DOE Announces HPSS Milestone - Todd Heer, Deputy Program Lead, Advanced Simulation and Computing (ASC) Facilities, Operations, and User Support (FOUS), announced that DOE High Performance Storage Systems (HPSS) eclipse one exabyte in stored data.

Atos Press Release - Atos boosts Météo-France’s data storage capacity to over 1 exabyte in 2025 to improve numerical modeling and climate predictions. Want to read more?

HPSS 9.2 Release - HPSS 9.2 was released on May 11th, 2021 and introduces eight new features and numerous minor updates.

Capacity Leader - ECMWF (European Center for Medium-Range Weather Forecasts) has a single HPSS namespace with over 679 PB spanning over 457 million files.

File-Count Leader - LLNL (Lawrence Livermore National Laboratory) has a single HPSS namespace with over 73 PB spanning 1.665 billion files.

Older News - Want to read more?
  • LLNL"
  • LANL"
  • NERSC"
  • ORNL"
  • Sandia"
  • IBM"
  • ANL"
  • Boeing"
  • BNL"
  • CEA"
  • CNES"
  • DWD"
  • DKRZ"
  • ECMWF"
  • PNNL
  • HLRS"
  • IU"
  • IITM"
  • IN2P3"
  • JAXA"
  • KEK"
  • KIT"
  • Met
  • MPCDF"
  • Meteo
  • NASA
  • NASA
  • NOAA
  • NOAA
  • NOAA
  • NOAA
  • Purdue"
  • SciNet"
  • SSC"
  • SLAC"
  • UTAS"
Home    |    About HPSS    |    Services    |    Contact us
Copyright 1992 - 2021, HPSS Collaboration. All Rights Reserved.