AWS's i4i instances with Nitro System v5 represent a breakthrough in cloud storage performance, and MayaScale delivers exceptional results on this platform—134 microsecond read and 186 microsecond write latency with shared storage that rivals direct-attached NVMe.
AWS i4i: The Nitro Advantage
Amazon Web Services' i4i instance family represents the latest evolution of their storage-optimized instances, powered by the revolutionary Nitro System v5. These instances provide AWS Nitro NVMe SSDs with hardware-accelerated I/O that dramatically reduces virtualization overhead.
While AWS's i4i instances provide excellent local storage, MayaScale takes performance further by pooling these NVMe resources across multiple instances using NVMe-over-Fabrics (NVMe-oF), delivering shared storage with latency that approaches direct-attached storage speeds.
Performance Breakthrough
In recent testing on AWS i4i.xlarge instances (4 vCPU, 1x AWS Nitro NVMe drive, 10 Gbps network), MayaScale achieved remarkable sub-millisecond latency performance that rivals direct-attached NVMe speeds while providing the flexibility of shared network storage.
Validated Performance Metrics
Peak Performance (Sub-1ms)
These metrics represent low queue depth (QD1-QD2) performance—the most important metric for real-world applications. Unlike synthetic benchmarks at high queue depths, low QD performance reflects what your applications actually experience.
Sequential Bandwidth Performance
In addition to exceptional random I/O performance, MayaScale delivers outstanding sequential throughput for large-block workloads:
Technical Architecture
Test Configuration
Our validation used AWS's i4i.xlarge instance (Basic tier):
- Instance Type: i4i.xlarge (4 vCPU, 32 GB RAM)
- Local Storage: 1x AWS Nitro NVMe SSD (1.875 TB capacity)
- Network: Up to 10 Gbps with Enhanced Networking (ENA)
- Configuration: Active-Active with RAID-1 replication
- Protocol: NVMe-over-TCP (NVMe-oF)
Why AWS Nitro Makes a Difference
The AWS Nitro System v5 offloads I/O processing to dedicated hardware, eliminating traditional hypervisor overhead. This results in:
- Direct NVMe access - Hardware-accelerated I/O path without hypervisor intervention
- Zero-copy DMA - Data moves directly between NVMe and network without CPU copies
- Dedicated I/O cores - I/O processing doesn't compete with application workloads
- Hardware encryption - Security without CPU overhead
This architectural advantage translates to 20-30μs lower latency compared to traditional virtualization, enabling MayaScale to achieve the exceptional 134μs read latency you see in our results.
Why These Numbers Matter
Traditional network storage typically delivers 1-5ms latency. Cloud block storage (like EBS io2) provides 2-10ms latency. MayaScale's sub-millisecond latency (0.134ms read, 0.186ms write) represents a 10-50x improvement over traditional solutions.
This performance level makes MayaScale suitable for applications previously requiring direct-attached storage:
- High-frequency trading systems - Where microseconds matter
- Real-time analytics - Processing data as it arrives
- NoSQL databases - Cassandra, MongoDB, ScyllaDB with ultra-low latency requirements
- In-memory databases - Redis, Memcached with persistent storage
- AI/ML training - Fast data loading for GPU workloads with Amazon SageMaker
Scaling Beyond the Basic Tier
While our testing focused on the i4i.xlarge (Basic tier) to demonstrate the lowest possible latency, AWS offers the entire i4i family for applications requiring higher IOPS and capacity:
- i4i.2xlarge - 8 vCPU, 1x 3.75TB NVMe, ~300K IOPS
- i4i.4xlarge - 16 vCPU, 1x 7.5TB NVMe, ~500K IOPS
- i4i.8xlarge - 32 vCPU, 2x 7.5TB NVMe, ~900K IOPS
- i4i.16xlarge - 64 vCPU, 4x 7.5TB NVMe, ~1.5M IOPS
- i4i.32xlarge - 128 vCPU, 8x 7.5TB NVMe, ~2.5M IOPS
The sub-millisecond latency characteristics remain consistent across all i4i instance sizes—it's a function of the AWS Nitro System and NVMe-oF protocol. Larger instances provide higher aggregate IOPS by pooling more NVMe devices.
Getting Started
MayaScale is available for AWS deployment via Terraform and will soon be available on AWS Marketplace. The solution includes:
- Automated deployment via Terraform modules
- Active-Active high availability with automatic failover
- NVMe-oF client drivers for Linux (Amazon Linux, RHEL, Ubuntu)
- Performance monitoring and CloudWatch integration
- Free trial available for evaluation
Visit our MayaScale on AWS page to learn more about deployment options, or contact our sales team to arrange a proof-of-concept on your AWS account.
The Future of Cloud Storage
The combination of AWS's Nitro System v5 and MayaScale's NVMe-oF architecture represents a fundamental shift in what's possible with cloud storage. Sub-millisecond shared storage is no longer theoretical—it's validated, production-ready, and available on AWS today.
Whether you're migrating databases from RDS to EC2 for better performance, running latency-sensitive applications, or building the next generation of cloud-native services, MayaScale on AWS i4i instances delivers the performance you need with the operational simplicity you expect from cloud infrastructure.
Ready to Experience 134μs Latency?
Deploy MayaScale on AWS in minutes. Get validated sub-millisecond performance with Active-Active high availability.
Contact Sales Download Free
