Back to Table of contents

Primeur weekly 2016-12-05

Exascale supercomputing

Hewlett Packard Enterprise demonstrates world's first Memory-Driven Computing architecture ...

Crowd computing

Einstein@home discovers new gamma-ray pulsar ...

Computing tackles the mystery of the dark universe ...

Quantum computing

Construction of practical quantum computers radically simplified ...

Researchers take first look into the 'eye' of Majoranas ...

More reliable way to produce single photons for quantum information imprinting ...

Focus on Europe

Deadline for ISC research paper submission extended to December 16 ...

PRACE to hold awards ceremony at Cineca for Summer of HPC 2016 ...

An open harbour for research data ...

New MareNostrum4 supercomputer to be 12 times more powerful than MareNostrum3 ...

Hardware

Lincoln Laboratory's supercomputing system ranked most powerful in New England ...

PolyU and Huawei jointly set up the first lab in optical communication and advanced computing system in Hong Kong ...

DDN Infinite Memory Engine burst buffer exceeds 1 TB per second file system performance for Japan's fastest supercomputer ...

Fujitsu develops in-memory deduplication technology to accelerate response for large-scale storage ...

Fujitsu announces start of operations for Japan's fastest supercomputer ...

Applications

Using supercomputers to illuminate the renaissance ...

IBM unveils Watson-powered imaging solutions for health care providers ...

New algorithm could explain human face recognition ...

Researchers use Stampede supercomputer to study new chemical sensing methods, desalination and bacterial energy production ...

Physicists spell 'AV' by manipulating Abrikosov vortices ...

BSC researchers to study the response of European climate to Arctic sea ice depletion ...

IBM and Pfizer to accelerate immuno-oncology research with Watson for Drug Discovery ...

Fujitsu offers deep learning platform with world-class speed and AI services that support industries and operations ...

The Cloud

Cloud Systems Management Software Market: Global Industry Analysis and Opportunity Assessment 2016-2026 ...

Juniper Networks simplifies Cloud transition for enterprises with carrier-grade routing and unified security for AWS marketplace ...

Fujitsu develops in-memory deduplication technology to accelerate response for large-scale storage


New in-memory deduplication technology.
5 Dec 2016 Kawasaki - Fujitsu Laboratories Ltd. has developed a high-speed in-memory data deduplication technology for all-flash arrays, which are large-scale, high-speed storage systems and use multiple flash devices such as solid-state drives. This technology enables the production of storage systems with up to twice the response speed when writing data, compared to previous methods.

In recent years, all-flash arrays have incorporated deduplication technology that consolidates duplicate data into one to write to a flash device, in order to utilize the limited capacity of flash devices. However, as the system connects to multiple flash devices through a network in order to search for duplicate data each time it writes data, and storage devices grow in capacity and increase in speed, a problem of lowered response speed during write operations arises.

Now Fujitsu Laboratories has developed a new method that can accelerate response speeds by executing deduplication after writing data. In addition, as data may be written to memory twice in some cases when processing is continued with the new method, thereby increasing communications volume and lowering overall processing performance, Fujitsu Laboratories has developed technology to automatically switch between the new method and the previous method, as operational conditions require.

This means that response speeds can be increased by up to two times, improving the response of virtual desktop services and reducing database processing times.

Details of this technology were announced at the 28th Computer Systems Symposium (ComSys2016), held November 28 at Hosei University in Tokyo.

In recent years, with the advance of virtualization and the use of Big Data in business, performance requirements for storage systems have been increasing. All-flash arrays, which can increase performance by using high-speed flash devices and connecting multiple devices with a high-speed network, have therefore become widely used.

Flash devices are more expensive than traditional hard drives, and can only be written to a limited number of times over their lifetime, so all-flash arrays incorporate deduplication functionality that consolidates data to write only once to the flash device, which can increase the actual volume and reduce the number of unnecessary write operations.

In order to utilize expensive flash devices effectively, in-memory deduplication technology was developed for all-flash arrays featuring deduplication functionality. However, because previous technology would institute a process to search for duplicate data before writing through a network that connects multiple flash devices, with storage systems that grow in capacity and increase in speed it created issues with longer response times when writing data due to the overhead required for search processing.

Fujitsu Laboratories has developed a technology that can improve response speed for in-memory deduplication up to two times.

Details of the technology are as follows:

1. New deduplication technology that improves response speed under low load

With the newly developed deduplication technology, the storage devices that make up the all-flash array cooperate to complete a write operation by temporarily storing the write data in open memory (the cache), improving response times. Then, while the server is preparing for the next write operation depending on the response from the memory, the technology searches for duplicate data, communicating between the storage devices in parallel, and the data is finally written to the solid-state drive after duplicates are eliminated. With this technology, response times for write operations can improve by up to two times when the storage system's network load is low.

2. Response time optimization technology actively switches between two deduplication methods

Deduplicating after responding improves response speed compared to previous methods when the storage system's network load is low, but when the load is high the network becomes congested. This reduces response times more than previous methods due to data writes across multiple devices taking longer.

With the newly developed response time optimization technology, of the two deduplication methods, the original method that writes after deduplication, for which the response time does not greatly vary, and the newly developed method, for which response times vary widely, the system automatically chooses the method with the shortest response time by calculating the expected value for the average response time, based on records measuring the actual response time.

With this technology, by automatically choosing the optimal method in response to the continually changing status of the system, it is possible to reduce system response time as a whole.

Using the newly developed technology, Fujitsu Laboratories was able to achieve a lowest latency of about half that of previous technology in fio benchmarks. As a result, Fujitsu Laboratories was able to increase response speed when writing data to an all-flash array by up to two times For example, in applications such as virtual desktops and data base processing that require high write speeds, as accessing small files occurs in enormous volumes, there are many duplications. In such situations, user applications on the service could be sped up, improving the user experience. In addition, by applying this system to back-end storage for operations databases, operations systems could be sped up, enabling further consolidation of IT infrastructure.

Fujitsu Laboratories will continue development of technologies to further accelerate all-flash arrays going forward, aiming to incorporate them into Fujitsu Limited's storage products from fiscal 2017 or later.
Source: Fujitsu

Back to Table of contents

Primeur weekly 2016-12-05

Exascale supercomputing

Hewlett Packard Enterprise demonstrates world's first Memory-Driven Computing architecture ...

Crowd computing

Einstein@home discovers new gamma-ray pulsar ...

Computing tackles the mystery of the dark universe ...

Quantum computing

Construction of practical quantum computers radically simplified ...

Researchers take first look into the 'eye' of Majoranas ...

More reliable way to produce single photons for quantum information imprinting ...

Focus on Europe

Deadline for ISC research paper submission extended to December 16 ...

PRACE to hold awards ceremony at Cineca for Summer of HPC 2016 ...

An open harbour for research data ...

New MareNostrum4 supercomputer to be 12 times more powerful than MareNostrum3 ...

Hardware

Lincoln Laboratory's supercomputing system ranked most powerful in New England ...

PolyU and Huawei jointly set up the first lab in optical communication and advanced computing system in Hong Kong ...

DDN Infinite Memory Engine burst buffer exceeds 1 TB per second file system performance for Japan's fastest supercomputer ...

Fujitsu develops in-memory deduplication technology to accelerate response for large-scale storage ...

Fujitsu announces start of operations for Japan's fastest supercomputer ...

Applications

Using supercomputers to illuminate the renaissance ...

IBM unveils Watson-powered imaging solutions for health care providers ...

New algorithm could explain human face recognition ...

Researchers use Stampede supercomputer to study new chemical sensing methods, desalination and bacterial energy production ...

Physicists spell 'AV' by manipulating Abrikosov vortices ...

BSC researchers to study the response of European climate to Arctic sea ice depletion ...

IBM and Pfizer to accelerate immuno-oncology research with Watson for Drug Discovery ...

Fujitsu offers deep learning platform with world-class speed and AI services that support industries and operations ...

The Cloud

Cloud Systems Management Software Market: Global Industry Analysis and Opportunity Assessment 2016-2026 ...

Juniper Networks simplifies Cloud transition for enterprises with carrier-grade routing and unified security for AWS marketplace ...