Exploring Summit: The newest intelligent supercomputer

Summit, the world’s fastest supercomputer running at Oak Ridge National Laboratory (ORNL), was designed from the ground up to be flexible and to support a wide range of scientific and engineering workloads. In addition to traditional simulation workloads, Summit is well suited to analysis and AI/ML workloads – it is described as “the world’s first AI supercomputer”. The use of standard components and software makes it easy to port existing applications to Summit as well as develop new applications. As pointed out by Buddy Bland, Project Director for the ORNL Leadership Computing Facility, Summit lets users bring their codes to the machine quickly, thanks to the standard software environment provided by Red Hat Enterprise Linux (RHEL).

Summit’s system is built using “fat node” building block concept, where each identically configured node is a powerful IBM Power System AC922 server which is interconnected with others via high-bandwidth dual rail Mellanox infiniband fabric, for a combined cluster of roughly 4,600 nodes. Each node in the system has:

SUMMIT SUPERCOMPUTER NODE COMPOSITION

The result is a system with excellent CPU compute capabilities, plenty of memory to hold data, high performance local storage, and massive communications bandwidth. Additionally, prominent use of graphical processing units (GPU) from Nvidia at the node architecture level provides robust acceleration platform for artificial intelligence (AI) and other workloads. All of this is achieved using standard hardware components, standard software components, and standard interfaces.

So why is workload acceleration so important? In the past, hardware accelerators such as vector processors and array processors were exotic technologies used for esoteric applications. In today’s systems, hardware accelerators are mainstream in the form of GPUs. GPUs can be used for everything from visualization to number crunching to database acceleration, and are omnipresent across the hardware landscape, existing in desktops, traditional servers, supercomputers, and everything in-between, including cloud instances . And the standard unifying component across these configurations is Red Hat Enterprise Linux, the operating system and software development environment supporting hardware, applications, and users across variety of environments at scale.

The breadth of scientific disciplines targeted by Summit can be seen in the list of applications included in the early science program. To help drive optimal use of the full system as soon as it was available, ORNL identified a set of research projects that were given access to small subsets of the full Summit system while Summit was being built. This enabled the applications to be ported to the Summit architecture, optimized for Summit, and be ready to scale out to the full system as soon as it was available. These early applications include astrophysics, materials science, systems biology, cancer research, and AI/ML.

Machine learning (ML) is a great example of a workload that stresses systems: it needs compute power, I/O, and memory to handle data. It needs massive number crunching for training, which is handled by GPUs. All of that requires an enormous amount of electrical power to run. The Summit system is not only flexible and versatile in the way it can handle workloads, it also withstands one of the biggest challenges of today’s supercomputers – excessive power consumption. Besides being the fastest supercomputer on the planet, it is equally significant that Summit performs well on the Green500 list – a supercomputer measurement of speed and efficiency which puts a premium on energy-efficient performance for sustainable supercomputing. Summit comes in at #1 in its category and #5 overall on this list, a very strong performance.

In summary, the fastest supercomputer in the world supports diverse application requirements, driven by simulation, big data, and AI/ML, employs the latest processor, acceleration and interconnect technologies from IBM, Nvidia and Mellanox, respectively, and shows unprecedented power efficiency for that scale of machines. Critical to the success of this truly versatile system is Linux, in Red Hat Enterprise Linux, as the glue that brings everything together and allows us to interact with this modern marvel.

New SAP HANA benchmark results help customers better plan their deployments

As traditional multi-tier enterprise software is adapting to new realities of cloud infrastructure, it also needs to make use of the latest advances in computational and hardware capabilities. Red Hat has been working with major ISVs and partners, like SAP, on digital transformation scenarios while simultaneously helping them to extract additional performance from their hardware with Red Hat Enterprise Linux.

As part of the quest for enhanced performance, the focus for database and analytics applications has been shifting to in-memory execution, a deployment model that SAP HANA is offering. In the future, that trend is likely to include even more complex designs that incorporate entire software frameworks for processing information in-memory, and that is where SAP Data Hub comes into play. As a result, last year Red Hat introduced an enhanced offering, Red Hat Enterprise Linux for SAP Solutions, that is designed to assist our customers in simplifying their adoption of Red Hat Enterprise Linux and to cater to various use cases they may have, including running SAP S/4 HANA.

To further aid customers and partners in planning, sizing and configuring their environments, SAP and Red Hat, along with other software and hardware partners, have historically used a suite of performance benchmarks. For traditional multi-tier deployments, the Sales and Distribution (SD) module became a “gold standard” for benchmarking across largest enterprises and small businesses alike. With a long history of collaboration with SAP and our mutual hardware OEM partners, like HPE and Dell EMC, among others, Red Hat is no stranger to delivering leading results on these benchmarks across multiple server sizes.

To demonstrate performance and provide additional scalability and sizing information for SAP HANA applications and workloads, SAP introduced the Business Warehouse (BW) edition of SAP HANA Standard Application Benchmark. Presently on version 2, this benchmark simulates a variety of users with different analytical requirements and measures the key performance indicator (KPI) relevant to each of the three benchmark phases defined as follows:

  1. Data load phase, testing data latency and load performance (lower is better)
  2. Query throughput phase, testing query throughput with moderately complex queries (higher is better)
  3. Query runtime phase, testing the performance of running very complex queries (lower is better)

As a result of close collaboration with our OEM partners, Red Hat Enterprise Linux (RHEL) was used in several recent publications of the above benchmark.

Specifically, processing 1.3 billion initial records (a popular dataset size) using a single Dell EMC PowerEdge R940xa server, demonstrated that running the workload on Red Hat Enterprise Linux could deliver the best performance across all three benchmark KPIs and outperform similarly configured servers (see Table 1).

 

Table 1. Results in scale-up category running SAP BW Edition for SAP HANA Standard Application Benchmark, Version 2 with 1.3B initial records

Phase 1

(lower is better)

Phase 2

(higher is better)

Phase 3

(lower is better)

Technology Release

Database Release

Red Hat Enterprise Linux 7.4 [1]

13,421 sec

10,544

99 sec

SAP NetWeaver 7.50 SAP HANA 1.0
SUSE Linux Enterprise Server 12 [2]

14,333 sec

6,901

102 sec

SAP NetWeaver 7.50 SAP HANA 1.0
Red Hat Enterprise Linux advantage

7%

53% 3%

 

Additionally, in a much larger dataset size of 5.2 billion initial records, Dell EMC PowerEdge R840 server running Red Hat Enterprise Linux also outscored similarly configured server on two out of three benchmark KPIs demonstrating better dataset load time and query processing throughput (see Table 2).

 

Table 2. Results in scale-up category running SAP BW Edition for SAP HANA Standard Application Benchmark, Version 2 with 5.2B initial records

Phase 1

(lower is better)

Phase 2

(higher is better)

Phase 3

(lower is better)

Technology Release

Database Release

Red Hat Enterprise Linux 7.4 [3]

74,827 sec

3,095

175 sec

SAP NetWeaver 7.50 SAP HANA 2.0
SUSE Linux Enterprise Server 12 [4]

84,744 sec

2,916

172 sec

SAP NetWeaver 7.50 SAP HANA 2.0
Red Hat Enterprise Linux advantage

13%

6% -1.75%

 

These results demonstrate Red Hat’s commitment to helping OEM partners and ISVs deliver high-performing solutions to our mutual customers, and showcase close alignment between Red Hat and Dell EMC that, in collaboration with SAP, led to the creation of certified, single-source solutions for SAP HANA. Available in both single-server and larger, scale-out configurations, Dell EMC’s solution is optimized with Red Hat Enterprise Linux for SAP Solutions.

Learn more: https://www.redhat.com/en/partners/dell and https://www.redhat.com/en/resources/red-hat-enterprise-linux-sap-solutions-technology-overview

 

Results as of July 30, 2018. SAP and SAP HANA are the registered trademarks of SAP AG in Germany and in several other countries. See http://www.sap.com/benchmark for more information.
[1] Dell EMC PowerEdge R940xa (4 processor / 112 cores / 224 threads, Intel Xeon
Platinum 8180M processor, 2.50 GHz, 64 KB L1 cache and 1024 KB L2 cache per core, 38.5 MB L3 cache per processor, 1536 GB main memory). Certification number #2018023
[2] FUJITSU Server PRIMERGY RX4770 M4 (4 processor / 112 cores / 224 threads, Intel Xeon
Platinum 8180 processor, 2.50 GHz, 64 KB L1 cache and 1024 KB L2 cache per core, 38.5 MB L3 cache per processor, 1536 GB main memory). Certification number #2018017
[3] Dell EMC PowerEdge R840 (4 processor / 112 cores / 224 threads, Intel Xeon
Platinum 8180M processor, 2.50 GHz, 64 KB L1 cache and 1024 KB L2 cache per core, 38.5 MB L3 cache per processor, 3072 GB main memory). Certification number #2018028
[4] HPE Superdome Flex (4 processor / 112 cores / 224 threads, Intel Xeon
Platinum 8180 processor, 2.50 GHz, 64 KB L1 cache and 1024 KB L2 cache per core, 38.5 MB L3 cache per processor, 3072 GB main memory). Certification number #2018025