Converged & Hyperconverged Infrastructure

 View Only

Resource Optimization with Hitachi HCI and vSAN HCI Mesh

By posted 03-09-2021 16:55

Hyperconverged Infrastructure (HCI) customers benefit from the flexibility of buy-as-you-grow economics provided by the scale-out architecture. Savings from the upfront investments in external storage infrastructure in exchange can be better utilized in other projects. However, certain scenarios require the independent scaling of compute and storage resources to better align with refresh cycles and resource consumption attributes of workloads. This independent scaling can be achieved using two different methods, depending on the use case, which we will discuss here today. As you will learn, Hitachi Vantara’s HCI platform (aka Unified Compute Platform HC, or short for UCP HC) is uniquely designed and optimized to offer both of these methods, providing utmost scalability, efficiency, and cost optimization.

Hitachi already provides a disaggregated storage architecture by supporting a design integrating HCI and external SAN Virtual Storage Platform (VSP) as detailed in the blog by my colleague Colin Gallagher here. Disaggregated HCI provides a high-performance, efficient and simplified solution for:
  1. Mission-critical apps with zero RPO needs and enterprise-grade snapshots where VSP storage with Global-Active Device technology fits the bill
  2. Apps requiring ultra-low latency where VSP shines with its proven scale and purpose-built drives
  3. Storage tiering to host data that efficiently meets information life cycle requirement 
  4. Full integration into VMware’s Storage Policy-Based Management (SBPM) framework with support for granular VM-level data services via VMware vSphere Virtual Volumes (vVols).  vVols and vSAN both share this common SPBM storage operational model, making the VSP storage an organic extension to Hitachi HCI
  5. One common interface to monitor, orchestrate, and lifecycle manage both the HCI compute cluster and the external storage resources, further optimizing operational efficiency
Many enterprises and federal departments with a hybrid infrastructure are benefitting from this scalable and flexible architecture.

Let us now discuss the second method. Recently introduced, vSAN HCI Mesh helps align workload requirements with the right resources by enabling independent UCP HC clusters to cross-leverage resources. UCP HC with HCI Mesh’s cross-cluster architecture complements existing disaggregated HCI solution and provides below customer benefits:

High-Performance Computing cluster:

Customers who deployed GPU-cluster with Hitachi UCP HC V225G for graphics-heavy apps may want to deploy a latency-critical workload that requires sub-millisecond latency. HCI Mesh now enables this GPU-optimized cluster to consume storage off another vSAN remote cluster that provides, say, additional high-performance storage using all-NVMe SSD drives. The unique approach to disaggregation brings together several independent clusters to deliver the best of both worlds, the most resource-efficient and cost-optimized design, and the highest-performing outcome, TOGETHER. Gone are the days when you had to worry about compromising between GPU cards, SSD drive count (IOPS or Capacity limits), and mismatched GPU to CPU ratios, this is HCI at its ultimate best.


Right-sized HCI resources:
HCI environments lend flexibility to start small and grow. However, over time they may need rebalancing as workloads grow or the new workloads are added causing either compute or storage constraints. It could mean inefficiency in software license and growing storage capacity may mean adding ESXi hosts which comes with additional license costs.

With the HCI Mesh architecture, customers can improve asset utilization and save on license costs by simply tapping into unused vSAN capacity from another vSAN cluster. The new architecture allows overutilized UCP clusters to remotely access vSAN datastores from underutilized UCP HC clusters hosted under the same vCenter. In addition, a compute-heavy cluster (think Quad-Socket nodes) can remotely mount a vSAN datastore from a capacity-optimized storage heavy UCP HC cluster (think 2U hybrid vSAN nodes), or from a performance-optimized UCP HC cluster (think all-NVMe all-Flash, further accelerated by Intel Optane technology) to balance and enhance resource utilization.

A few key use cases for UCP HC Mesh architecture are below
1. OLTP: UCP HC Mesh architecture will provide flexibility to deploy log files and OS on a computing cluster while hosting the database on another high-performance storage cluster to deliver fast query response. This architecture will help avoid the virtualization and database license costs for scale-out of the existing capacity-constrained cluster. 
2. VDI: Environments that have more persistent desktop users usually require storage scaling at a different rate than linked-clone desktops. UCP HC Mesh allows better resource allocation to cater to a variety of user-profiles without piling on license costs.  In other cases, vSAN storage capacity could also grow over time to increase the capacity requirements for the Windows Servers hosting various shared folders used for user documents, profiles, etc., and the Mesh design provides again the same benefit here.

In-memory DB and Analytics: Apps that are constrained by limited memory and storage can be supported with UCP HC meshed design. Customers can run more VMs hosting in-memory analytics per vSphere hosts (in what we’ll call the “compute cluster”) by remotely mounting a high-performance vSAN datastore from another HC cluster (consider this the “storage cluster”, more NVMe drives, more capacity, more IOPS) without having to scale out the existing “compute cluster” and add licensing costs. The performance increase can be dramatic, and the cost savings significant as well (can license 4-way, 8-way CPUs on the compute cluster for example, and only license single-CPU or dual-CPUs on the storage cluster!)

New disaggregated storage design will be useful in large and small HCI environments alike. Both methods of scaling independently compute and storage resources have their clear and distinct use cases, as we explained earlier, and both can play a major part in your hyperconverged infrastructure!

What do you think of the new HCI Mesh architecture and its unique use cases? Share your thoughts in the comment below.




05-19-2022 14:35

Very Informative !!

05-02-2022 02:04


04-26-2022 13:04

Good Read