Written by Aaron Melin, Sr. Director, Engineering, Liqid
Hypervisor-agnostic,
Composable Solutions Deliver Adaptive, Software-defined Hardware Optimization
and Increased Disk Group Capacity for VMware's VSAN and More
As artificial intelligence and machine
learning (AI+ML) become status quo for both data center automation and higher
order applications, the hyperconverged infrastructure (HCI) architectures are
headed for a breaking point. Unable to scale to meet the uneven demands of
these highly virtualized compute environments, additional, tightly bundled hardware
must be deployed and virtualized. But with limited ability to disaggregate data
center resources, these increasingly sprawling HCI systems can quickly become
uneven, with some hardware sitting idle while other devices are taxed to their
limits.
Further compounding the matter, as AI+ML
solutions are utilized to study and optimize virtualized data centers, the
ability to share GPU, FPGA and other accelerator technologies necessary for
these data-intensive applications is physically limited by hardware configuration.
Without the ability to pool these high-value resources through software and
adapt for changes in unpredictable, data-centric workloads, the benefits of
virtualization can be significantly diminished.
CDI for
HCI
To address the challenges associated with the
evolving virtualized data center, software-defined, composable, disaggregated
infrastructure (CDI) solutions and services automate, orchestrate, and compose
resources at the bare-metal level. IT users can significantly increase data agility,
capacity, and bandwidth in virtualized, HCI environments for which lightning
fast, adaptive data manipulation is a necessity. Pools of disaggregated GPU,
FPGAs, CPUs, NVMe SSD, and storage class memory extension technologies are
leveraged to create balanced systems that can adapt on demand or through
automation to surges in data activity.
Hypervisor-agnostic composable infrastructure
is optimized for VMware's VSAN and other leading providers. The ability to
create multiple servers on demand with composable software reduces the number
of licenses required for virtualization deployments. Once a task is completed,
those resources can be repurposed either on demand, or through policy or
AI+ML-driven automation.
IT users can also deploy composable Intel Optane
technology, users can create multiple virtual caching drives on a single, high
capacity add-in-card (AIC) for ultra-high data performance, enabling the
ability to scale to more disk groups per server. Liqid composable software also
ensures complete fault tolerance, instantly transferring data from failed
compute nodes to other available disaggregated hardware.
Multi-fabric
and Multifaceted
The marketplace's leading composable
infrastructure solutions are multi-fabric, and can be deployed up and down the
hardware stack via PCIe to achieve maximum performance, or across Ethernet and
Infiniband fabrics to pool and share resources across distance.
The ability to share across multiple fabrics
also means that NVMe-over-Fabric and even GPU-over-Fabric operations can be
performed with maximum speed and efficiency for virtualized clusters.
Disaggregated resources can be pooled and shared regardless of where they
physically reside, delivering another way to share once-static resources
through software, and reducing the need to make additional hardware purchases
to achieve the same results.
For virtual desktop infrastructure
deployments, the move to composable infrastructure provides agility for
hardware that, again, can see extremely different demands depending on the time
of day. IT users can push GPU resources toward VDI applications in the morning
to deal with the dreaded 9:00 AM ‘boot storm,' then redeploy those resources in
tandem with different devices for analytics, AI+ML, and other high-value data
tasks. Resources no longer need to be reserved in a silo to manage remote
workers for 30 minutes in the morning.
Composable infrastructure expands the promise
of composability, allowing IT users to create and maintain a more efficient
data center infrastructure to meet the crushing and inconsistent data
performance requirements of today's highly virtualized compute environments, as
well as prepare for emerging, high value applications.
##
About the Author
Aaron Melin, Sr. Director, Engineering, Liqid
Aaron Melin is an experienced director of software development with a demonstrated history of success in engineering for composable infrastructure solutions, data accelerators, networking & intelligent fabrics, high performance computing, and overall data center architectures. Before joining Liqid, he held engineering leadership positions at companies such as Pivot3, Dot Hill Systems, Brocade Communications Systems, and other top IT providers. Aaron earned his bachelor's degree in Computer Engineering from Northeastern University, and he lives in the greater Denver area with his family, where they enjoy spending time outdoors and fostering German Shepherds.