» Current | 2018 | 2017 | 2016 | Subscribe

Linley Newsletter

TidalScale Inverts Server Virtualization

June 5, 2018

Author: Bob Wheeler

We seldom write about software companies, but TidalScale offers a software-based alternative to the proprietary hardware traditionally employed in scale-up servers. First, forget everything you know about virtual machines (VMs). Whereas traditional server virtualization divides one physical server into many small VMs, TidalScale enables one massive virtual machine running across many physical servers. Its approach creates a virtual scale-up server running a single operating-system instance on hardware composed of commodity x86 servers. This approach benefits applications requiring more memory than a single commodity server provides, such as in-memory databases, analytics, and scientific computing.

Other large-scale nonuniform memory architectures (NUMAs) rely on high-speed low-latency interconnects to minimize the cost of remote-memory access. TidalScale’s innovation is its use of machine learning to dynamically characterize resource consumption and then redistribute resources to minimize communications between physical nodes. In many cases, its HyperKernel moves a process to a new node rather than moving memory contents to the original node.

The distributed HyperKernels create a single memory image, with each node’s physical memory operating as a level-four (L4) cache. More than a decade ago, AMD and Intel enabled hierarchical virtual-memory structures to handle hypervisor-based virtualization. Those same extensions make TidalScale’s HyperKernel completely transparent to the guest operating system (OS). In fact, the HyperKernel doesn’t distinguish between OSs and applications, neither of which require modification.

Many innovative processor and system architectures are solutions in search of a problem. TidalScale’s approach, however, appears well suited to in-memory computing, which has exploded thanks to big data. Scale-out software, such as Hadoop, introduces performance-robbing overhead and can require time-consuming data-set partitioning. By turning virtualization on its head, TidalScale expands memory in a new way, obviating the need for physical scale-up servers.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Fall Processor Conference 2018
Covers processors and IP cores used in embedded, communications, automotive, IoT, and server designs.
October 31 - November 1, 2018
Hyatt Regency, Santa Clara, CA
Register Now!
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »