ServiceNow’s move to run its AI-first workflow platform on Azure Ultra Disk is a quiet but consequential shift: the company has validated that managed block storage on Azure can meet the sub-millisecond latency, high‑IOPS, and global scale needs of a mission‑critical SaaS platform that serves 85% of the Fortune 500, and has used that capability to deploy Ultra Disk across 14 Azure regions—enabling a new-region rollout in roughly six weeks where it once took months.
Background / Overview
ServiceNow is best known for unifying IT service management, IT operations, and enterprise workflows into a single cloud platform. Over the last several years the company has doubled down on an
AI-first strategy—introducing RaptorDB (its next‑generation HTAP database), AI Control Tower, and embedded agent/ecosystem capabilities that place low‑latency database access and predictable storage performance at the heart of user experience and automated agents. Those changes increased pressure on the underlying storage substrate: throughput, predictable latency tails, and global availability became gating factors for both performance and regional expansion. Faced with this reality, ServiceNow evaluated cloud storage options against exacting criteria—
sub‑millisecond latency, high IOPS, robust throughput, and worldwide availability—and selected Azure Ultra Disk Storage as the only managed disk offering that met all requirements. That decision is documented in Microsoft’s customer story and is reflected in ServiceNow’s hybrid configuration where on‑prem or direct‑attached NVMe remains part of the stack for some workloads, while Ultra Disk provides the managed, scalable cloud foundation for global SaaS operations. This article explains the technical rationale, validates key claims against public documentation, highlights what the choice means in practice for enterprise SaaS and database operators, and offers a sober look at both the strengths and risks of anchoring a high‑performance AI/SaaS platform on managed cloud block storage.
Why storage matters for enterprise AI and workflow platforms
Modern, agentic AI and HTAP databases change the performance equation for SaaS platforms in three critical ways:
- Low end‑to‑end latency is mandatory because agents and UX components rely on near‑instant query results to meet human expectations.
- Mixed I/O profiles (high random IOPS for transactional workloads plus sustained throughput for analytics and model checkpoints) require storage that can deliver both tail‑latency consistency and high bandwidth.
- Global scale and regional availability are operational imperatives—customers expect local latency SLAs, data residency control, and rapid regional provisioning.
ServiceNow’s platform stitches workflows, live analytics, and agent orchestration together. That union amplifies the cost of storage‑induced latency: a slow query or long tail can ripple into degraded agent decisions, failed automations, and poor human experience. The practical upshot is simple—when database and storage are the bottleneck, optimization of compute alone is insufficient. ServiceNow’s selection criteria reflect that reality.
What Azure Ultra Disk offers (specs and validation)
Azure Ultra Disk is Microsoft’s high‑end managed block storage offering aimed at data‑intensive, latency‑sensitive workloads. Its key technical characteristics—relevant to ServiceNow’s decision—include:
- Configurable IOPS and throughput per disk with very high ceilings (provisioned IOPS up to the disk and VM limits). The Ultra Disk family supports a wide range of IOPS/throughput configurations to match workload needs.
- Designed for consistent, low latency suitable for databases such as PostgreSQL, MySQL/MariaDB, and enterprise SAP HANA workloads. Azure’s docs describe Ultra Disks as “high throughput, high IOPS, and consistent low latency” storage for VMs.
- A managed disk model with built‑in encryption, availability guarantees, and the ability to dynamically scale IOPS/throughput without rebooting VMs in many scenarios; these operational characteristics simplify lifecycle and compliance work.
Microsoft’s product blog and documentation provide the raw engineering baseline. ServiceNow’s case study adds the real‑world layer: ServiceNow used Ultra Disk to reach
performance parity with its direct‑attached NVMe systems for the databases that power its AI Platform, and to standardize on a managed, multi‑region deployment model spanning 14 Azure regions. Those are vendor‑level claims corroborated by the published Ultra Disk spec sheet and Azure documentation, but they also deserve cautious interpretation—benchmarks run by platform vendors often depend on workload tuning and the precise VM/disk sizing used.
RaptorDB, MariaDB, and database considerations
ServiceNow’s platform relies on two major database surfaces:
- MariaDB (legacy transactional workloads and customer migrations), and
- RaptorDB, ServiceNow’s proprietary, PostgreSQL‑based HTAP database designed to fuse transactional and analytical workloads for the Now Platform. ServiceNow publishes performance guidance and benchmarks for RaptorDB—with claims of significant speedups across queries, transactions, and reporting.
Why this matters: HTAP databases amplify the need for a storage layer with both excellent transactional IOPS and the ability to sustain high analytic throughput. RaptorDB’s integrated column‑store indexes, parallel processing, and other HTAP features reduce ETL needs and shift more real‑time work onto the platform—but they also increase the importance of a storage substrate that maintains consistent latency under concurrent, mixed workloads.
ServiceNow’s approach has been hybrid: retain custom Dell NVMe systems where it makes sense, and validate Ultra Disk against those baselines through a three‑month benchmarking and validation process before declaring parity for production workloads. That hybrid stance provides both performance continuity and a migration path to managed cloud operations.
Operational impact: speed of regional expansion and deployment cadence
One of the most tangible outcomes ServiceNow reports is a dramatic acceleration in geographic expansion. According to the Microsoft customer story, ServiceNow now deploys Ultra Disk across 14 Azure regions and can bring a new region online in about six weeks—a significant improvement over the months‑long provisioning cycles previously required when relying solely on private datacenter procurement and provisioning. Operational benefits include:
- Faster time‑to‑market for new regional SaaS availability.
- Less capital tied up in long‑lived hardware purchases; the cloud model enables hardware refreshes and VM/instance class changes with lower friction.
- Simplified lifecycle operations, including snapshot-based backups and managed recovery paths (ServiceNow mentions exploring Azure Snapshot backup and premium SSD v2 options going forward).
These gains are consistent with cloud migration economics—hyperscalers can shorten procurement and commissioning times—but they require operational discipline: configuration automation, repeatable benchmarking, and cross‑team runbooks to avoid configuration drift and to preserve the tight SLA targets required by enterprise customers.
Strengths of the approach
- Managed performance at scale: Azure Ultra Disk delivers a managed storage fabric with fine‑grained control over IOPS and throughput, enabling predictable database performance without complex local storage management. Microsoft’s docs and Ultra Disk announcements confirm sub‑millisecond latency targets and high IOPS/throughput configurations designed for databases and transaction‑heavy workloads.
- Operational agility: Moving to a managed cloud foundation reduces hardware lifecycle risk, accelerates region provisioning, and allows ServiceNow to pivot to newer VM classes and storage features without the capital and time overhead of on‑prem upgrades. The company explicitly states this shift enabled quicker access to new hardware and virtualization options.
- Integrated DB support: Ultra Disk supports the database technologies ServiceNow needs, including MariaDB and the company’s RaptorDB, removing one of the technical blockers that often stalls cloud migrations—storage incompatibility or unpredictable latency tails.
- Hybrid continuity: ServiceNow’s hybrid model—combining its existing direct‑attached NVMe platforms with Ultra Disk—lets the company incrementally validate workloads, giving the engineering teams space to tune and certify Azure SKUs for production without wholesale cutovers.
Risks, caveats, and questions that deserve scrutiny
No infrastructure decision is free. ServiceNow’s public case study is a strong endorsement of Azure Ultra Disk, but several operational and strategic risks remain that IT leaders and architects should consider carefully.
- Vendor claims vs. independent verification
- ServiceNow and Microsoft report performance parity with on‑prem NVMe and cite 99.99% availability—claims that are credible but are ultimately vendor‑provided. Independent third‑party benchmarks against the same workloads would be ideal to validate parity under production-like multi‑tenant stress. Treat the published figures as vendor‑validated but not universally proven for every workload.
- Cost and price predictability
- Ultra Disk’s configurable IOPS/throughput model provides flexibility, but it can also create complexity for cost forecasting. High‑performance I/O provisioning at global scale is expensive; careful cost modeling and monitoring are needed to prevent unexpected spend escalation when autoscaling or provisioning for peak loads.
- Cloud provider coupling and resilience
- Relying on a single cloud provider for the storage foundation increases operational coupling risk. While Azure offers broad global footprint and enterprise support, customers must still design for provider outages, cross‑region failover, and data residency/regulatory constraints. The operational benefits of managed storage must be balanced with multi‑region and multi‑provider resilience strategies where required.
- Database migration complexity
- Moving a global SaaS database topology to a managed disk changes fault‑domains, replication topologies, and failover semantics. ServiceNow’s three‑month validation and hybrid strategy mitigate risk, but others attempting similar moves should expect non‑trivial migration engineering—especially for mission‑critical transactional metadata.
- Tail latency and mixed workload interference
- Ultra Disk promises consistent low latency, but mixed HTAP workloads can expose tail‑latency problems at scale (e.g., large analytic scans interfering with transactional I/O). ServiceNow’s integration of RaptorDB and workflow fabrics will need careful I/O QoS and workload scheduling to prevent degraded performance under peak analytical activity.
- Benchmark and measurement transparency
- ServiceNow describes a benchmarking process that involves simulating load and analyzing CPU, memory, and latency results. For the industry, broader transparency—published methodologies, reproducible test cases—would help customers evaluate claims more objectively.
Practical guidance for architects and IT leaders
ServiceNow’s experience provides a practical blueprint for enterprises considering a similar migration or performance re‑platforming. The following steps condense the lessons into an actionable sequence:
- Inventory and classify workloads
- Identify transactional vs analytic workloads, tail‑latency sensitivity, and I/O profile (random vs sequential).
- Pilot with representative production workloads
- Run your realistic workload patterns (not synthetic microbenchmarks) against Ultra Disk configurations; measure tail latency, not just averages.
- Validate database compatibility and tuning
- Test replication topologies, failover behavior, and consistency guarantees under simulated failures.
- Build an automated validation pipeline
- Use infrastructure as code to reproduce environments, run benchmark suites, and collect traceable metrics.
- Model cost and create guardrails
- Provision budgets and set monitoring alerts for IOPS/throughput spending; use autoscaling or tiering strategies to move non‑critical workloads to cheaper tiers.
- Maintain hybrid fallback during ramp
- Keep the existing on‑prem or direct‑attached NVMe capacity during a staged migration: this preserves an escape hatch and avoids blast radius in case of performance surprises.
- Plan for multi‑region resilience
- Design replication and failover across regions, and document RTO/RPO expectations; regulatory controls may require region‑local replication or data residency constraints.
This pragmatic checklist reflects both ServiceNow’s practice (three‑month validation, hybrid deployments) and general cloud migration best practices.
What this means for the enterprise SaaS market
ServiceNow’s public endorsement of Azure Ultra Disk is significant for three reasons:
- It signals hyperscalers can now credibly host extremely latency‑sensitive, high‑IOPS SaaS platforms that historically demanded co‑located NVMe hardware.
- It accelerates the normalization of HTAP and agentic AI SaaS stacks running on managed cloud primitives—reducing the custom hardware premium for enterprise platform vendors.
- It increases the importance of storage engineering as a first‑class concern for SaaS product teams—decisions about disk class, IOPS allocation, VM SKU, and regional placement now materially affect product SLAs and speed of expansion.
For cloud providers, the bar has been raised: enterprises will expect not only raw VM and GPU capacity but also block storage fabrics that provide predictable microsecond‑scale latency at global scale.
Strengths and weaknesses summarized
- Strengths
- Predictable, configurable IOPS and throughput for databases.
- Managed operations, encryption, and integration with Azure IaaS services.
- Faster regional provisioning and the ability to pivot to new VM/storage classes.
- Support for modern database workloads including MariaDB and ServiceNow’s RaptorDB.
- Weaknesses / Risks
- Vendor claims need independent benchmarking; “parity with NVMe” should be validated on a per‑workload basis.
- Potential for higher operating cost without aggressive storage governance.
- Increased provider coupling and the need for careful DR/multi‑region planning.
- HTAP workloads require careful I/O isolation to avoid contention and tail‑latency exposures.
Final analysis: opportunity balanced with pragmatic caution
ServiceNow’s adoption of Azure Ultra Disk illustrates how enterprise SaaS vendors can marry high‑performance storage with the practical benefits of managed cloud operations. The technical fit—Ultra Disk’s low latency and configurable IOPS, combined with ServiceNow’s RaptorDB HTAP engine—creates a compelling architecture for real‑time AI workflows and global SaaS scale. Microsoft’s published customer story documents the deployment across 14 regions, the six‑week regional onboarding cadence, and the company’s three‑month validation process—an operational model other vendors will study closely. That said, the move is not a one‑size‑fits‑all guarantee. The claims of parity with direct‑attached NVMe are meaningful but should be treated as
vendor‑validated results and validated independently by any organization considering a similar migration. Enterprises should combine careful workload classification, realistic benchmarking, cost governance, and multi‑region resilience planning before committing fully.
The practical implications are clear: top‑tier managed block storage is now a credible foundation for enterprise‑grade AI and workflow platforms in the public cloud. For IT leaders, the takeaway is straightforward—if your platform needs predictable, low‑latency I/O at global scale, Ultra Disk is a technology worth piloting, but pilot it the right way: with production‑representative workloads, rigorous measurement of tail latency, and a plan that balances speed of expansion with cost and resilience controls.
ServiceNow’s case is a template—not an automatic decision. It shows the path, the wins, and the traps. For organizations advancing their AI and workflow platforms, the next step is deliberate measurement: match your workload, run the tests, and make the migration with both ambition and operational guardrails in place.
Source: Microsoft
https://www.microsoft.com/en/customers/story/25578-servicenow-azure-load-balancer/