Lead Image © sdecoret, 123RF.com

Lead Image © sdecoret, 123RF.com

Computational storage that supports storage operations

Smart Assistant

Article from ADMIN 79/2024
ScaleFlux delivers performance gains of 50 percent and more with computing power built directly into the network card to relieve the burden on the CPU.

ScaleFlux competes with NVMe-based drives by offering not only store but compute functions, as well. However, relieving the load on the CPU and ensuring higher bandwidth and lower latency comes at a greater price and begs one question: Is it worth the effort?

Where Is the Bottleneck?

Every server has a bottleneck that limits its performance somewhere, but the exact whereabouts of this bottleneck tends to differ because the manufacturers of the individual components have been playing a cat-and-mouse game for years. New CPUs usually come with new chipsets that enable higher bandwidths for the connected devices and RAM, which often makes the storage devices the bottleneck. The storage device manufacturers then follow suit and launch faster devices so that the network suddenly becomes the performance limit. The network in particular has seen major innovations in recent years, including the now widely available 400Gb Ethernet, which has quickly passed the buck back to storage devices.

The situation can become particularly critical when components interact, especially if these components happen to be exposed to a heavy compute load (e.g., database nodes, servers with running Hadoop or Redis instances, or members of an Elasticsearch cluster). Several factors come together. In particular, individual steps such as compressing and indexing data or cleaning up data records that are no longer needed require serious bandwidth on the connected storage medium on the one hand and high compute power on the part of the CPU on the other.

In the worst case, this leads to a single system slowing down an entire setup, such as when the central database can no longer respond to incoming queries or cannot respond quickly enough. Neither users nor administrators like this one bit, because – especially on the end-user side – using a service in this scenario is a pain.

The problem is not new, which

Use Express-Checkout link below to read the full article (PDF).

Buy this article as PDF

Express-Checkout as PDF
Price $2.95
(incl. VAT)

Buy ADMIN Magazine

Get it on Google Play

US / Canada

Get it on Google Play

UK / Australia

Related content

comments powered by Disqus
Subscribe to our ADMIN Newsletters
Subscribe to our Linux Newsletters
Find Linux and Open Source Jobs

Support Our Work

ADMIN content is made possible with support from readers like you. Please consider contributing when you've found an article to be beneficial.

Learn More”>


		<div class=