My company is replacing an aging Hitachi storage array with a new HPE system. The existing environment was designed by employees who have since left, and my manager has mainly maintained it for the last four or five years. I'll be responsible for configuring the new array and connecting it to our clustered Hyper-V hosts, but I haven't previously built a SAN from scratch. What should I consider for the initial design, including RAID or data-protection layout, storage pools, LUN and volume design, networking, multipathing, and capacity planning? We'll also need to migrate workloads from the old array, so advice on running both systems alongside each other and minimizing downtime or migration risk would be appreciated. I'm trying to identify the major pitfalls before we begin.
5 Answers
Measure the existing environment before designing the replacement. Record capacity, peak and average IOPS, throughput, latency, workload concurrency, and growth rate. Test representative sequential and random reads and writes if possible. Then size the new array around performance, usable capacity, failure behavior, and the organization’s recovery requirements rather than simply filling it with the largest possible volumes. Follow the exact best-practices guide for the HPE model; an MSA, Nimble-derived platform, and other HPE systems can have very different provisioning and RAID recommendations.
Plan the migration as a separate project. Confirm that both arrays can be connected at the same time, verify compatibility and support for the chosen migration method, and test backups and restores before moving production workloads. Migrate a noncritical workload first, monitor latency and host paths, and schedule the remaining moves with a rollback plan. Keep the old array intact until every workload has been validated and you have a documented recovery procedure.
Before anyone can recommend a layout, you need to identify the exact HPE model, whether the hosts use Fibre Channel or iSCSI, the number of Hyper-V hosts, available interfaces, capacity requirements, workload and I/O patterns, and whether compression or deduplication will be used. SAN platforms differ significantly, so generic advice can be misleading. For business-critical workloads, purchasing the vendor’s implementation service is worthwhile. Stay involved, take notes, and ask the engineer why each design choice is being made. It gives you both a correctly supported deployment and useful training.
There are many ways to create a configuration that works initially but causes performance or recovery problems later. Watching an experienced storage engineer build it is far safer than relying only on a manual.
Don’t assume that a simple LUN layout or a particular RAID level is universally correct. Modern arrays often abstract disk groups and protection through pools or distributed layouts, while some platforms have specific rules about volume sizes, workload types, snapshots, thin provisioning, and deduplication. Use the vendor’s design guide and have the proposed configuration reviewed before committing it. Also document array settings, switch configuration, host initiators, zoning or VLANs, MPIO settings, firmware versions, and monitoring alerts so the environment is maintainable after deployment.
Treat storage connectivity as a dedicated infrastructure design. Use redundant storage switches or fabrics where appropriate, separate storage networks and VLANs from normal server traffic, and configure MPIO correctly on every Hyper-V host. Make sure the array and hosts have enough independent links and that the switches, transceivers, paths, and controller ports are actually redundant. More connections are not automatically better; an incorrect or unbalanced design can perform worse than fewer properly configured paths. Validate the final result with path-failure tests and realistic workload benchmarks, not just a basic disk-speed test.
The important part is end-to-end balance: host adapters, switch bandwidth, controller ports, queue depths, MPIO policy, and the array’s actual capabilities all need to match. A large number of links alone does not guarantee more throughput.
It’s also worth getting that recommendation and service approved in writing. If management declines professional installation, make sure the risk and your recommendation are documented.