Case Study

Benchmarking Block Storage: Nirvana ABS vs AWS EBS

April Wong

— 3 min read
Benchmarking Block Storage: Nirvana ABS vs AWS EBS

Benchmarking Nirvana ABS vs AWS EBS gp3 and io2 with ClickBench and FIO

TL;DR

  • Nirvana ABS is ~10–14× faster than AWS gp3 on cold-read analytical workloads
  • Nirvana ABS performs similarly to io2 (within 2× on ~98% of queries)
  • Nirvana ABS delivers ~20K sustained IOPS per VM in real mixed workloads
  • Nirvana ABS is ~18% cheaper than gp3 and ~8× cheaper than io2
    → Strong default for analytics, blockchain, and data-heavy workloads

Block storage is a foundational layer for modern infrastructure. From analytics pipelines to blockchain nodes, performance at the storage layer directly impacts latency, throughput, and overall system reliability.

To understand how different storage options behave under real workloads, we benchmarked Nirvana Accelerated Block Storage (ABS) against two widely used AWS options: EBS gp3 (general-purpose) and EBS io2 (premium high-IOPS).

We used two complementary approaches:

  • ClickBench to measure real analytical SQL performance under cold-read conditions
  • FIO to evaluate raw storage behavior across realistic I/O patterns

ClickBench: Cold-Read Analytics

ClickBench executes 43 OLAP queries while dropping the OS page cache before each run, forcing all reads to come directly from storage. This exposes true storage performance in scenarios where caching cannot hide bottlenecks.

Under these conditions, ABS delivered ~10–14× faster performance than gp3 on I/O-heavy queries. The gap is most visible on large scans and aggregations, where gp3 shows significant runtime expansion while ABS maintains stable execution times.

Compared to AWS io2, ABS remained highly competitive:

  • ~77% of queries ran within 1.5× of io2
  • ~98% of queries ran within 2×

In practice, ABS demonstrates io2-class analytical behavior across most workloads.

FIO: Real Infrastructure Behavior

To isolate storage performance, we ran FIO tests across a 16-VM cluster using the full production data path (VM → CSI → iSCSI → storage backend).

Key results:

  • ~20K sustained IOPS per VM under mixed (80/20) workloads
  • ~3.5 ms latency at 4K writes
  • ~408–430 MB/s sequential throughput per VM

These patterns align closely with real-world systems such as databases, indexing pipelines, and blockchain state storage.

Price-Performance

We normalized costs to a common scenario: a 1TB volume delivering ~10K sustained IOPS.

Monthly cost:

  • ABS: $95.7
  • AWS gp3: $116.9
  • AWS io2: $778

ABS is ~18% cheaper than gp3 and ~8× cheaper than io2, while delivering comparable performance in analytical workloads.

Takeaway

Across both benchmarks, ABS delivers a strong price-performance profile:

  • Eliminates gp3 cold-read performance cliffs
  • Matches io2 behavior across most workloads
  • Provides significantly better cost efficiency

For teams running analytics, blockchain infrastructure, or data-heavy systems, ABS offers a more predictable and cost-effective storage foundation.

Nirvana: Modern Cloud for Real-time Workloads

Ultra-fast Block Storage with High IOPS, powering blockchain, AI and real time systems.

Learn more at Nirvana Labs

Nirvana Cloud| Pricing | Blog | Docs | Twitter | Telegram | LinkedIn

Read more