openEuler Summit Sharing: Migration Practices of iSoftStone HPC based on openEuler

On November 9-10, the 2nd Operating System Summit & openEuler Summit was held in Beijing to gather the strength of the whole industrial chain, and jointly promote the development of the basic software industry. iSoftStone as a community co-constructor is the co-organizer of the Summit, and delivered the themed speech of “Migration Practice of HPC based on openEuler Operating System” at the Compatibility Forum on Nov.10.

From multiple dimensions such as promotion and application of the root technology based on Huawei openEuler + Kunpeng platform, and SPARK performance improvement with openEuler KAE, iSoftStone comprehensively shared its outstanding practices of migration and adjusting in openEuler operating system.


HPC-WAN creates high-timeliness and high-performance meteorological application solutions

In the age of big data, the meteorological service application field keeps expanding, and meteorological data processing becomes more important. However, the data size of meteorological applications to be processed is large with a high requirement for timeliness. Performance is the crucial requirement of meteorological application for the platform. HPC-WAN migration practice is to migrate the meteorological solution to openEuler+ Kunpeng computing platform, and bring performance priority of openEuler+ Kunpeng in HPC scenarios into full play. iSoftStone compiles through BiSheng Compiler, replaces MPI with HyperMPI, and regulates OS performance to finally realize the normal operation of meteorological HPC components on the openEuler+Kunpeng platform. The overall performance is superior to a common platform.

Based on the root technology of the Huawei openEuler+Kunpeng platform, the performance of HPC-WAN solutions is higher than the common platform (X86), and the requirement of the meteorological industry for high performance of computing platform is effectively solved. Through HPC-WAN migration practice, iSoftStone drives the promotion and application of Huawei basic software root technology in the industry.

SPARK makes openEuler create big data high-efficient processing components

SPARK is an open-source big data processing component, and supports a group of rich advanced tools, including sparksql for SQL. The compress modes of current open-source Spark have pros and cons. GZIP mode has no enabling KAE, high CPU utility and slow compression speed. The optimization enables openEuler acclib in Spark, and improves Spark performance through software and hardware acceleration pool of openEuler. iSoftStone adds KAEzip calling interface of openEuler in Spark, enables gzip interface and unzipping pool configuration; and applies scala development language for KAE hardware acceleration and enabling. The Spark SQL inquiry performance after final optimization increases 5+ times, and HiBench performance tests improve 40%+.

The AKE acceleration ability based on Huawei openEuler drives SPARK big data processing efficiency to improve 5+ times; iSoftStone enables openEuler optimization practice through SPARK, and supports enabling and application of Kunpeng KAE acceleration in open-source software.

As the root technology of the basic software, the operating system is the key force to boost digital reform. As a reliable partner for enterprise digital transformation, iSoftStone helps numerous industries carry out practices of digital transformation, and continuously concerns and supports the development of the operating system industry. iSoftStone, as a participant and contributor of openEuler, actively participates in community cooperation and communication, ceaselessly improves technical capacity and accumulates a large amount of experience about acceleration pool and application migration direction, and makes a continuous contribution to the ecological development of the openEuler operating system.

Contact Us










Zip code:100193

Telephone:+86 105874 9000

Fax:+86 105874 9001

Address:East Bldg. 16, Courtyard #10, Xibeiwang East Road, Haidian Dist., Beijing 100193, China