Short answer: A server and storage refresh BOQ should start from workload data, not from the configuration you bought five years ago. Measure current CPU, memory, IOPS and capacity utilisation, add realistic growth, decide what stays on-premises versus moves to cloud or colocation, and only then specify hardware. The three line items most often missed are usable capacity after overheads, licensing under current vendor models, and the support response time that actually matches your downtime tolerance.
Key takeaways
Specify from measured utilisation and growth, not from the previous generation's spec sheet.Usable capacity is what matters. Raw capacity minus RAID, overhead and free-space headroom is a much smaller number.Licensing has changed enough in recent years that it can exceed the hardware line. Price it early.Support response time, not warranty length, determines what an outage costs you.Decide the destination mix — on-premises, colocation, cloud — before you buy anything.When a refresh is actually due
Warranty or support expiry, especially where the vendor no longer offers extension at a sensible priceFirmware and OS support end, which quietly turns into a compliance and audit findingPerformance complaints that trace back to storage latency or memory pressure rather than the applicationCapacity pressure, where free space or consolidation ratios leave no room for growth or failoverPower, cooling or rack constraints in a server room that was never designed for today's loadA pending project — ERP upgrade, virtualisation change, AI workloads, a new site — that the current estate cannot absorbStep 1 — Measure before you specify
Collect at least two to four weeks of data, covering a month-end or peak period:
CPU utilisation per host, average and 95th percentileMemory allocated versus consumed, including ballooning or swappingStorage: used and free capacity by tier, IOPS, throughput, and read-write latencyGrowth rate of data over the last 12 to 24 monthsVirtual machine count, sizes, and how many are idle or oversizedBackup window length and whether it currently completesNetwork throughput per host and any bottleneck at the uplinkRight-sizing against measured data regularly removes a tier of hardware from the BOQ, or redirects that budget into flash storage and memory where it does more good.
Step 2 — Decide the destination mix first
Keep on-premises: latency-sensitive systems, large stable datasets, applications tied to local equipment or plant systems.Move to colocation: workloads that need better power, cooling and physical security than a server room offers, while staying on your own hardware. See Data Centre Colocation in Chennai.Move to cloud: variable or seasonal workloads, development and test environments, and anything already being modernised. See Azure Migration Services in Chennai.Retire or consolidate: servers running a single legacy application nobody owns. Every refresh should remove something.The split decides how much hardware you actually buy, so settle it before the BOQ is drafted rather than after quotes arrive.
Step 3 — Architecture choice, in plain terms
Three-tier (servers, SAN, switches): familiar, granular scaling of compute and storage separately, more components to manage and integrate.Hyperconverged (HCI): compute and storage in the same nodes, simpler operations and scaling in node units, licensing per node to watch.Hybrid with cloud bursting or DR in cloud: steady-state on-premises, recovery or overflow in a cloud region. See Azure Disaster Recovery for On-Premises Applications.There is no universally correct answer. The deciding factors are usually the size of your IT team, how predictable growth is, and whether storage and compute grow at the same rate.
Step 4 — Specify storage by behaviour, not by terabytes
All-flash or NVMe for databases, virtualisation and anything latency-sensitiveHybrid or capacity tiers for archives, file shares and backup landing zonesUsable versus raw: state the usable figure after RAID or erasure coding, formatting overhead, snapshots, and the free-space headroom the platform needs to performIOPS and latency targets per workload class, written into the BOQ as acceptance criteriaData services included: snapshots, replication, deduplication, compression, encryption at rest — and whether each needs a separate licenceGrowth path: can you add shelves or nodes, and at what granularity and costStep 5 — Price licensing and support honestly
Operating system licensing, counted the way the vendor counts it today, including the core minimums per serverHypervisor and management licensing, with current subscription terms and multi-year renewal pricingDatabase licensing, which is frequently the largest and most easily mis-specified itemBackup software licensing by capacity, socket or workloadHardware support: response time, onsite spares, and whether parts are stocked in Chennai or shipped from another cityWho performs firmware updates and health checks after go-live, and whether that is includedAsk for renewal pricing in year four at the time of purchase. A refresh that looks competitive at year one can become expensive at the first renewal.
Step 6 — Do not forget the environment
Rack units required, including room for growth and for the transition period when old and new run togetherPower draw per node, PDU capacity, and whether existing UPS and generator cover the new loadCooling capacity and airflow direction in the roomNetwork ports and speeds required, including switch uplinks and any upgrade to the top-of-rack switchesCabling, labelling and patching workPhysical security, access control and environmental monitoring in the server roomBOQ checklist to send with your RFP
Measured utilisation summary and growth assumption used for sizingWorkload list with performance and availability requirementsNode or server specification: sockets, cores, memory, boot mediaStorage specification with usable capacity, IOPS and latency targetsNetwork adapters, port speeds and switch requirementsRedundancy: power supplies, paths, failure domains, N+1 assumptionsLicence list with term, metric and year-four renewal pricingBackup target and retention designDR design and recovery targets, with a test planSupport level, response time and spare parts locationInstallation, migration and cutover servicesAcceptance test criteria and a performance baseline after go-liveDecommissioning, data sanitisation and asset disposalProject plan with downtime windows and rollback pointsCommon mistakes
Specifying the new estate as a copy of the old one, including its oversized virtual machines.Quoting raw capacity instead of usable capacity, so the array is full a year early.Leaving licensing until the end, then discovering it exceeds the hardware budget.Buying next-business-day support for systems that cannot be down for a day.Ignoring power, cooling and rack space until delivery day.No acceptance testing, so performance claims are never verified.No plan for the old equipment, its data, or its disposal.Frequently asked questions
How often should servers be refreshed?
Commonly every four to six years, driven by warranty, support status and workload fit rather than the calendar alone. Storage is often refreshed on a similar cycle, with capacity expansions in between.
Should we move to the cloud instead of refreshing?
For some workloads, yes. The honest comparison is total cost over the refresh period for both options, including licensing, operations and connectivity — and a workload-by-workload decision rather than an all-or-nothing one.
Is hyperconverged infrastructure cheaper than a traditional SAN?
Sometimes, depending on scale, licensing and how independently compute and storage need to grow. Compare over the full term, including node-based licensing and the cost of the smallest expansion step.
How much capacity headroom should we plan for?
Enough to absorb measured growth across the planned life, plus free space for the platform to perform and for failure domains to rebuild. Specify usable capacity after all overheads rather than raw.
What should be in the acceptance test?
Measured IOPS and latency against the stated targets, failover of a node or path, backup and restore of a representative workload, and a performance baseline that is recorded for future comparison.
What happens to the old hardware?
Decide upfront: trade-in, redeployment to test or DR, or disposal. For disposal, insist on documented data sanitisation and a certificate — it will be asked for in your next security or ISO 27001 audit.
Do we need GPUs in this refresh?
Only if you have identified AI or visualisation workloads. If you do, size them separately — see GPU Server Quotation in India.
Plan your refresh with Layots
Layots Technologies plans and delivers server, storage and virtualisation refreshes for businesses in Chennai, Bangalore and Hyderabad: utilisation assessment, right-sizing, vendor-neutral BOQ preparation, licensing review, migration and cutover, acceptance testing, and managed operations afterwards.
Send us your current estate and what the next three years look like for your business. A solution architect reviews the requirement and comes back with a sized design and a costed BOQ, usually within one business day.
Book a 30-minute virtual meeting with a solution architect — request your free IT assessmentCall +91 99623 95939WhatsApp the team for a quick requirement discussionRelated reading: Choosing an IT Infrastructure Partner in Chennai · Backup and Disaster Recovery for Chennai Businesses · Nutanix Cloud Clusters on Azure BareMetal