HPE Data Management Framework System Architecture Overview ...€¦ · Overview HPE Data Management...
Transcript of HPE Data Management Framework System Architecture Overview ...€¦ · Overview HPE Data Management...
HPE Data Management Framework system architecture overview
Contents Section 1. Overview .................................................................................................................................................................................................................................................................. 2 Section 2. Major solution components ................................................................................................................................................................................................................... 3 Section 3. System configuration examples ......................................................................................................................................................................................................... 5
Lustre File System with object/cloud back-end storage ................................................................................................................................................................. 5 Lustre File System with tape back-end storage ...................................................................................................................................................................................... 6 HPE XFS all-flash or disk file system with cloud/object back-end storage .................................................................................................................... 7 HPE XFS all-flash or disk file system with tape back-end storage ......................................................................................................................................... 7 HPE XFS all-flash or disk file system with DMF zero watt storage and tape back end ...................................................................................... 8
Section 4. Summary ................................................................................................................................................................................................................................................................. 9
Technical white paper
Technical white paper Page 2
Section 1. Overview HPE Data Management Framework (DMF) provides a large-scale, tiered storage and data management platform specifically engineered to manage and protect the petabytes of structured and unstructured content generated by highly scalable and extremely dynamic HPC, data analytics, and artificial intelligence workflows.
HPE DMF combines industry-leading traditional, cloud, and cost-optimized storage devices with software-defined storage management features to create a powerful yet cost-effective data management environment. As such, HPE DMF is a major enabling technology component in support of data-intensive compute environments reducing storage costs, simplifying long-term data management, and improving the levels of data protection for critical projects.
HPE DMF enables the intelligent blending of high-performance storage located close to the point of data ingest and compute with cost-managed, high-capacity storage based on multiple technologies that may include object storage, cloud storage, and library-based tape. The HPE DMF-based approach allows selected data to be stored on low-cost, capacity-optimized storage media instead of or in addition to the premium-tier technology such as SSDs or fast hard drives while also meeting the target levels of storage performance.
This white paper provides an overview of the HPE DMF deployment architecture along with information on sample system configurations and data throughput capabilities.
Technical white paper Page 3
Section 2. Major solution components An HPE Data Management Framework (DMF) deployment consists of core elements that include:
• File system namespaces that are accessed by users and applications
– HPE DMF supports direct integration with Lustre (version 2.10 or later) and HPE XFS file systems.
• HPE DMF servers for data management and movement
– Core HPE DMF server nodes that ingest file system events (change log); store metadata in a scale-out, distributed Cassandra database; process DMF policies and queries via Spark in-memory analytical engine; and coordinate data movement request processing.
– HPE DMF data mover nodes that perform the actual transfer of data between the user-accessible file systems and the back-end storage where HPE DMF-managed data is protected and replicated. Data mover nodes also perform periodic data validation activities to protect organizations from data loss due to media failure or silent corruption. Finally, data mover nodes are responsible for large-scale, nondisruptive data migration and movement activities between the prior version of media and the newer technology.
• Back-end storage technologies
– Object and cloud storage based on the de facto S3 API protocol standard—HPE DMF can tier data to any S3-compliant cloud or object storage technology.
– DMF zero watt storage based on high-performance, high-density JBOD storage enclosures such as the HPE D6020 (70 drives, standard rack) and HPE D8000 (106 drives, deep rack)—This solution provides extremely low storage costs along with reduced operating expenses based on drive spin-down and power control that saves on both power and cooling costs.
– Tape-based storage leveraging industry-standard, high-end tape libraries and drive technologies—HPE DMF provides validated integration with virtually all major tape libraries from vendors including Hewlett Packard Enterprise, Spectra Logic, Oracle (StorageTek), and IBM. Support for LTO, Oracle, and IBM tape drive technology is also provided.
Figure 1 shows a typical DMF deployment architecture with each of the technology areas called out.
Figure 1. HPE DMF solution elements
HPE DMF v7 architecture elements
HPE DMF core servers and metadata
DMF parallel data movers
HPC cluster
Fabric
Lustre or HPE XFS parallel file system
Tape drives
Cloud or object storage
SAN
DMF zero watt storage
File System/namespace
Back-end storage technologies
HPE DMF data management and mover servers
Technical white paper Page 4
Within HPE DMF configurations, the HPE DMF core server and data mover nodes are typically implemented using HPE DL360 and DL380 server configurations.
The 1U HPE ProLiant DL360 Gen10 server is a dual-socket server using Intel® Xeon® Scalable processors, supporting up to 3 TB of memory in 24 DIMM slots, and three expansion slots when populated with two processors. Many options are available for network connectivity using either FlexibleLOM or PCIe network adapters.
Figure 2. HPE ProLiant DL360 Gen10 server
The 2U HPE ProLiant DL380 Gen10 server is a dual-socket server using Intel Xeon Scalable processors, supporting up to 3 TB of memory in 24 DIMM slots, and up to eight expansion slots when populated with two processors. Many options for network connectivity are supported using FlexibleLOM and PCIe network adapters.
Figure 3. HPE ProLiant DL380 Gen10 server
Technical white paper Page 5
Section 3. System configuration examples Lustre File System with object/cloud back-end storage Figure 4 provides high-level system configuration examples for HPE DMF server counts and network connectivity in cases where an HPE DMF solution would be integrated with a Lustre file system environment—and with a cloud or object back-end configuration for capacity-optimized data storage. In this example, Ceph is shown as the object storage back end, but any S3 API-compatible object/cloud storage technology could be substituted.
Figure 4. HPE DMF with Lustre and cloud/object storage
Figure 5 provides performance information of the large configuration shown in the Figure 4 illustration with five HPE DMF core servers and four HPE DMF data movers.
Figure 5. HPE DMF deployment example with Lustre and cloud/object storage
Lustre
IB/OPA
Ethernet
Ceph
Configuration elements‒ Three (total) HPE DMF nodes that will perform
the core/database and data mover functions
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Two dedicated data mover nodes
Configuration elements‒ Five dedicated HPE DMF 7 core/database nodes‒ Four dedicated data mover nodes‒ Multiple Lustre file systems
Lustre
IB/OPA
Ethernet
Ceph
DB
Lustre
IB/OPA
Ethernet
Ceph
DB
Lustre
Mover MoverMover
POC and small configuration Typical configuration Large configuration
5 Lustre file systems
InfiniBand
100GbE
Mover 4
Mover 2
Mover 3
Mover 1
RGW 1
RGW 2
RGW 3
RGW 4
RGW 5
RGW 6
SUSEEnterprise
Storage
Target:10 GB/s–20 GB/s
DMF 2
DMF 3
DMF 4
DMF 1
DMF 5
Technical white paper Page 6
The goal of this configuration was to deliver 10 GB/s to 20 GB/s of HPE DMF driven data movement between Lustre and Ceph. The Figure 6 shows that this objective was met with each of the data movers providing between 4 GB/s to 5 GB/s of throughput for an aggregate system bandwidth of over 18 GB/s.
Figure 6. HPE DMF data movement throughput for four data mover configuration
Lustre File System with tape back-end storage Figure 7 provides high-level system configuration examples for HPE DMF server counts and network connectivity in cases where an HPE DMF solution would be integrated with a Lustre file system environment—and with a tape-based back-end configuration for capacity data storage.
Figure 7. HPE DMF with Lustre and tape-based storage
Lustre
IB/OPA
Tape SAN
Lustre
IB/OPA
Tape SAN
DB
Lustre
IB/OPA
Tape SAN
DB
Lustre
Mover MoverMover
Tape library Tape library Tape library
Lustre Lustre Lustre Lustre
POC and small configuration Typical configuration Large configuration
Configuration elements‒ Three (total) HPE DMF nodes that will perform
the core/database and data mover functions
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Two dedicated data mover nodes
Configuration elements‒ Five dedicated HPE DMF 7 core/database nodes‒ Four dedicated data mover nodes‒ Multiple Lustre file systems
Technical white paper Page 7
HPE XFS all-flash or disk file system with cloud/object back-end storage Figure 8 provides high-level system configuration examples for HPE DMF server counts and network connectivity in cases where an HPE DMF solution would be integrated with an HPE XFS file system environment based on either hard-disk or all-flash storage—and with a cloud or object back-end configuration for capacity data storage. In this example, Ceph is shown as the object storage back end, but any S3 API-compatible object/cloud storage technology could be used.
Figure 8. HPE DMF with HPE XFS and cloud/object-based storage
HPE XFS all-flash or disk file system with tape back-end storage Figure 9 provides high-level system configuration examples for HPE DMF server counts and network connectivity in cases where an HPE DMF solution would be integrated with an HPE XFS hard disk or all-flash file system environment—and with a tape-based back-end configuration for capacity data storage.
Figure 9. HPE DMF with HPE XFS and tape-based storage
RAID
IB/OPA/SAN
S3 Ethernet
Ceph
Configuration elements‒ Three (total) HPE DMF core/database and data
mover nodes‒ One combined HPE XFS metadata server and
NAS edge server
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Two dedicated HPE XFS metadata servers‒ Two dedicated HPE DMF data mover nodes‒ Two dedicated NAS edge servers
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Three dedicated HPE XFS metadata servers‒ Four dedicated data mover nodes‒ Six dedicated NAS edge servers‒ Multiple RAID or flash file systems
Flash
NAS access
DB DB DBMDS DB Mover NAS
NAS access
S3 Ethernet
Ceph
RAID
IB/OPA/SAN
FlashMDS
DB Mover NAS
NAS access
S3 Ethernet
Ceph
RAID
IB/OPA/SAN
FlashMDS
POC and small configuration Typical configuration Large configuration
RAID
IB/OPA/SAN
Tape SAN
Tape library
Flash
NAS access
DB DB DBMDS DB Mover NAS
NAS access
Tape SAN
RAID
IB/OPA/SAN
FlashMDS
DB Mover NAS
NAS access
RAID
IB/OPA/SAN
FlashMDS
Tape library
Tape SAN
Tape library
POC and small configuration Typical configuration Large configuration
Configuration elements‒ Three (total) HPE DMF core/database and data
mover nodes‒ One combined HPE XFS metadata server and
NAS edge server
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Two dedicated HPE XFS metadata servers‒ Two dedicated HPE DMF data mover nodes‒ Two dedicated NAS edge servers
Configuration elements‒ Three dedicated HPE DMF core/database nodes‒ Three dedicated HPE XFS metadata servers‒ Four dedicated data mover nodes‒ Six dedicated NAS edge servers‒ Multiple RAID or flash file systems
Technical white paper Page 8
HPE XFS all-flash or disk file system with DMF zero watt storage and tape back end Figure 10 provides high-level system configuration examples for HPE DMF server counts and network connectivity in cases where an HPE DMF solution would be integrated with an HPE XFS hard disk or all-flash file system environment—and with a combined zero watt storage (ZWS) and tape-based back-end configuration for capacity data storage.
The DMF zero watt storage component can act as a high-performance cache facilitating fast recalls of warm data without mounting the tapes. DMF ZWS can be assigned a power budget that will limit the number of disks that can be powered and spun up at a given time.
Figure 10. HPE DMF with HPE XFS, zero watt storage, and tape
RAID
IB/OPA/SAN
TapeSAN
Tapelibrary
Configuration elements‒ Three combined HPE DMF 7 database, library
server, data mover, and Open Vault storage nodes‒ One combined HPE XFS metadata server and
NAS edge server
Configuration elements‒ Three dedicated HPE DMF 7 Database nodes‒ Two dedicated HPE XFS metadata servers‒ Two combined data movers and Open Vault
storage nodes‒ Two dedicated NAS edge servers
Configuration elements‒ Three dedicated HPE DMF 7 database nodes‒ Three dedicated HPE XFS metadata servers‒ Four combined data movers and Open Vault
storage nodes‒ Six dedicated NAS edge servers
Flash
NASaccess
DB DB DBMDS DB Mover NAS
NASAccess
TapeSAN
RAID
IB/OPA/SAN
FlashMDS
DB Mover NAS
NASaccess
RAID
IB/OPA/SAN
FlashMDS
Tapelibrary
TapeSAN
Tapelibrary
ZWS ZWSZWS
POC and small configuration Typical configuration Large configuration
Technical white paper
Share now
Get updates
© Copyright 2019 Hewlett Packard Enterprise Development LP. The information contained herein is subject to change without notice. The only warranties for Hewlett Packard Enterprise products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. Hewlett Packard Enterprise shall not be liable for technical or editorial errors or omissions contained herein.
Intel Xeon is a trademark of Intel Corporation in the U.S. and other countries. Oracle is a registered trademark of Oracle and/or its affiliates. All other third-party marks are property of their respective owners.
a00072412ENW, May 2019
Section 4. Summary The configuration and performance examples provided in this document represent just a sample of the environments where HPE DMF may be leveraged. In preparation for HPE DMF deployments, HPE engages with customers in HPE DMF workshops that are intended to provide system architects, administrators, and users with the information on how HPE DMF will integrate into their particular storage environment and job workflow. The output of these sessions will typically include a target system architecture, sizing, and throughput document that will support the development of a final HPE DMF system configuration.
Figure 11. HPE DMF data management transformation workshop
Learn more at hpe.com/storage/dmf
Customer benefitsOne-day visual and interactive session allows HPE experts to focus on customers’ needs for next-generation data management and help embark on a transformational journey to accelerated job execution, improved data leverage, and reduced administrative burden.
Create a data management plan and road map
‒ Select and analyze HPC and AI workflows and data use cases
‒ Identify opportunities for flash-based acceleration, storage cost reductions, and improved data protection
‒ Assess data characteristics: data readiness, requirements
Scope
‒ Get started on a data management project quickly‒ Align business, data, and IT operations teams‒ Explore opportunities, priorities, and select relevant use cases
‒ Identify dependencies, data sources, level of readiness‒ Define a high-level road map for intelligent data management strategy