MayaScale Validated Performance Tiers on AWS

MayaScale delivers validated sub-millisecond performance across 4 performance tiers on AWS, from 204K to 1.35M IOPS. See how i4i and i3en instances enable ultra-low latency shared storage with Active-Active HA.

Four Validated Performance Tiers on AWS

We've completed comprehensive performance validation of MayaScale across AWS's storage-optimized instance families. The results demonstrate consistent sub-millisecond latency and validated IOPS performance across four distinct tiers, giving you flexibility to choose the right price/performance balance for your workload.

All testing was conducted using SNIA-compliant FIO benchmarks between October and November 2025, with results validated across multiple deployment configurations. Each tier has been tested for both zonal and cross-AZ (regional) deployments with Active-Active high availability.

Performance Tier Overview

Tier	Instance	Write IOPS	Read IOPS	Latency	Write BW	Read BW	Cost/mo
Basic	i4i.xlarge	57,000	204,000	~200 µs	223 MB/s	797 MB/s	$416
Standard	i3en.2xlarge	135,000	346,000	~200 µs	527 MB/s	1.35 GB/s	$876
High	i3en.6xlarge	368,000	992,000	~250 µs	1.44 GB/s	3.87 GB/s	$2,628
Ultra	i3en.12xlarge	528,000	1,350,000	~250 µs	2.06 GB/s	5.27 GB/s	$5,256

Note: All metrics are for zonal deployments. Cross-AZ deployments have ~10% write IOPS reduction due to network latency.

Basic Tier: Ultra-Low Latency Leader

Configuration

• Instance: i4i.xlarge (4 vCPU, 32 GB RAM)
• Storage: 1× AWS Nitro NVMe SSD (1.875 TB)
• Network: Up to 10 Gbps Enhanced Networking
• Architecture: AWS Nitro System v5

Why AWS Nitro Makes a Difference

The i4i Basic tier achieves the lowest latency of any MayaScale tier across all clouds thanks to AWS's Nitro System v5. Hardware-accelerated I/O processing eliminates traditional hypervisor overhead, delivering 20-30μs lower latency than conventional virtualization.

Best For

• Development and testing environments
• Applications requiring absolute minimum latency
• Small-scale production workloads (< 200K IOPS)
• CI/CD pipelines and build systems
• Ultra-low latency proof-of-concept deployments

Standard Tier: Best Price/Performance

Configuration

• Instance: i3en.2xlarge (8 vCPU, 64 GB RAM)
• Storage: 2× NVMe SSDs (2.5 TB each, 5 TB total)
• Network: Up to 25 Gbps
• Capacity: 5 TB usable with RAID-1

Price/Performance Analysis

Standard tier offers the best price/performance ratio across all MayaScale tiers on AWS. At $6.49 per 1K write IOPS, it's 53% better value than Ultra tier ($9.95/1K IOPS) while maintaining the same sub-millisecond latency characteristics.

Best For

• Production databases (PostgreSQL, MySQL, MongoDB)
• Web application file storage
• General-purpose shared storage
• VMware datastores (moderate VM count)
• Container persistent volumes (Kubernetes/EKS)

High Tier: Performance Sweet Spot

Configuration

• Instance: i3en.6xlarge (24 vCPU, 192 GB RAM)
• Storage: 2× NVMe SSDs (7.5 TB each, 15 TB total)
• Network: 25 Gbps
• Optimal QD: QD16×12 (192 outstanding I/Os)

The Performance Sweet Spot

High tier represents the optimal balance between performance and cost for most enterprise workloads. It delivers nearly 1M read IOPS while maintaining sub-millisecond latency, all at a lower cost per IOPS than Ultra tier.

Best For

• High-performance databases requiring sub-1M IOPS
• Analytics workloads (Spark, Presto, Trino)
• ML training pipelines
• Video editing and rendering
• Large-scale VMware environments

Ultra Tier: Maximum Capacity & Read Performance

Configuration

• Instance: i3en.12xlarge (48 vCPU, 384 GB RAM)
• Storage: 4× NVMe SSDs (7.5 TB each, 30 TB total)
• Network: 50 Gbps
• Optimal QD: QD16×20 write, QD48×48 read

Understanding the Ultra Tier Bottleneck

Comprehensive testing revealed that Ultra tier write performance is limited by AWS i3en NVMe device characteristics, not network or CPU. The 4× local NVMe drives deliver 635K IOPS raw (79% of AWS's 800K spec), which translates to 528K IOPS after RAID-1 overhead—an excellent 83% overall efficiency for HA storage.

Read performance, however, scales beautifully to 1.35M IOPS, making Ultra tier ideal for read-heavy workloads.

Best For

• Read-heavy workloads requiring >1M IOPS
• Content delivery and caching systems
• Large-capacity requirements (>15 TB)
• Data warehousing with read-optimized queries
• Combined high-performance + high-capacity needs

Note: Ultra tier has diminishing returns for write-heavy workloads. Consider High tier for better write IOPS cost efficiency.

Technical Architecture

Active-Active High Availability

All MayaScale tiers on AWS use Active-Active architecture with server-side RAID-1 replication. Both storage nodes serve I/O concurrently, delivering:

• 2× read performance: Both nodes serve read requests simultaneously
• Sub-second failover: Automatic failover in under 1 second
• Zero data loss: Synchronous replication ensures consistency
• Cross-AZ deployment: Nodes in different availability zones for resilience

NVMe-over-Fabrics Protocol

MayaScale uses NVMe-over-TCP (NVMe-oF) to pool local NVMe storage across instances. This cloud-native protocol delivers:

• Sub-millisecond latency: Direct NVMe access over standard Ethernet
• Zero kernel patches: Standard Linux NVMe-oF drivers
• Works with ENA: Leverages AWS Enhanced Networking
• iSCSI compatibility: Also supports legacy iSCSI initiators

Testing Methodology

All performance metrics were validated using:

• Tool: FIO with libaio engine
• Block size: 4KB for random I/O, 128KB for sequential
• Runtime: 30 seconds per test (SNIA-compliant)
• Queue depths: QD1, QD2, QD4, QD8, QD16, QD32, QD48, QD64
• Concurrency: Multiple numjobs (threads) tested
• Protocol: Direct I/O (bypassing page cache)

Cost and Value Analysis

Tier	Cost/mo	Write IOPS	$/1K Write IOPS	Read IOPS	$/1K Read IOPS	Value Score
Basic	$416	57K	$7.30	204K	$2.04	Good
Standard	$876	135K	$6.49	346K	$2.53	✅ Best
High	$2,628	368K	$7.14	992K	$2.65	Good
Ultra	$5,256	528K	$9.95	1.35M	$3.89	⚠️ Expensive

Tier Selection Recommendations

• Choose Basic if you need absolute minimum latency (134μs) and moderate IOPS
• Choose Standard for the best price/performance ratio (most workloads)
• Choose High for performance-critical applications requiring sub-1M IOPS
• Choose Ultra only for read-heavy workloads needing >1M IOPS or >15TB capacity

Getting Started

MayaScale is available for AWS deployment via Terraform modules and will soon be available on AWS Marketplace. Choose your deployment method:

Terraform Deployment

Infrastructure-as-Code deployment with full control over configuration:

module "mayascale" {
  source = "github.com/zettalane/mayascale-terraform//aws"
  
  performance_tier = "aws-zonal-standard"  # Basic, Standard, High, or Ultra
  region          = "us-east-1"
  availability_zones = ["us-east-1a", "us-east-1b"]
}

AWS Marketplace (Coming Soon)

One-click CloudFormation deployment with:

• Automatic billing through AWS account
• Pre-configured Active-Active HA setup
• Integrated with AWS Console
• Production-ready in under 15 minutes

Conclusion

MayaScale's validated performance across all four AWS tiers demonstrates the maturity and consistency of NVMe-over-Fabrics technology on public cloud. From the ultra-low latency Basic tier (134μs, $416/mo) to the high-capacity Ultra tier (1.35M IOPS, 30TB, $5,256/mo), every tier delivers sub-millisecond shared storage with Active-Active high availability.

Whether you're migrating databases from RDS, building latency-sensitive applications, or deploying VMware workloads on AWS, MayaScale provides validated, production-ready performance with transparent pricing and deployment automation.