Benchmarking Hyperscalers: Aiven for ClickHouse Delivers More for Less

Aiven is proud to introduce our new pricing plans for Aiven for ClickHouse! Depending on your chosen region and plan, you can now double your compute power for the same price. On the other hand, if reducing costs is your priority, it's now possible to lower your total cost by up to 30% for the same compute power, depending on how you use the service. But how did we achieve this? This article explains how, and why Aiven’s nodes are the best for your ClickHouse workloads. Choosing Aiven offers peace of mind. We've already evaluated all the hyperscaler options for you, allowing you to focus on your business.

This benchmarking wasn't just about picking a few new VMs. It was a comprehensive research mission. We're excited to share our journey, our findings, and more importantly, how this translates into a faster, more flexible, and cost-effective Aiven for ClickHouse service for you.

The Challenge: Navigating the instance maze for ClickHouse

Choosing the right compute instance for your workload is complex. Hyperscaler cloud providers offer a wide array of instance types with different CPU types and generations, varying memory-to-CPU ratios, and many storage options like network-attached storage, local instance store, and attached SSDs. All these options can be intimidating, making instance selection confusing.

By over-provisioning your instances, you over-spend. By under-provisioning, you risk losing your users to slow queries and customer incidents. Compounding this challenge, the number of available instance types grows annually, and the optimal type for your workload may not yet be available in your region.

To ensure that we offer the best price-performance available, we analyzed all the available options to find the sweet spot for Aiven for ClickHouse.

Our methodology for Benchmarking: the ClickBench standard

We chose ClickBench for our benchmark because it is a robust standard specifically designed for an analytical DBMS like ClickHouse.

Here’s a glimpse into our methodology:

1. The Dataset: A 100-million-row "hits" dataset, reflecting common ClickHouse practices.
2. The Queries: 43 ClickBench queries to test various analytical workloads.
3. Hot vs. Cold Runs: Each query was run multiple times, simulating both initial (cold) and subsequent (hot) data access scenarios.
4. Performance Score: We calculated a geometric mean of normalized scores, providing a single, balanced performance score for each VM, where a lower score indicates better performance.
5. Pricing: We compared instances in AWS us-west-2 and GCP us-central-1, including at least 1TB of storage for a fair baseline. For VMs with attached SSDs, we used that storage for benchmarks; for others, we used EBS gp3 on AWS or Hyperdisk on GCP.

The Contenders: A Look at AWS & GCP Instance Families

We cast a wide net, focusing on promising instance families from both AWS and GCP, including those with Intel, AMD, and ARM (Graviton) processors, and various RAM-to-CPU ratios and storage options. A key decision was prioritizing network-attached storage, like AWS EBS and GCP Hyperdisk, for its flexibility and dynamic disk sizing capabilities. This is a game-changer for Aiven for ClickHouse.
A key consideration was the performance of SSDs that we offer today, compared to the flexibility of network-attached storage. Because SSDs are limited in size and complex to upgrade, we chose network-attached storage so we could offer dynamic disk sizing (DDS).

AWS Instances Compared

General Purpose (EBS-backed)

Storage Optimized (Attached SSDs)

GCP Instances Compared

General Purpose (Hyperdisk/Persistent Disk)

Compute Optimized (Hyperdisk/Persistent Disk)

The New Champions: Faster, More Flexible, and Cost-Effective

Our rigorous benchmarking led to clear winners, translating directly to improved price performance.

(Remember, the lower, the better!)

For AWS: The m8g Series with EBS Shines

• ARM's Dominance: Across the board, ARM instances (m8g, i8g) consistently outperformed their intel x86 counterparts (m7i, i7ie) in price and performance. The m8g is cheaper and faster - a significant advantage for ARM instances.
• A Crucial Decision: m8g (EBS) vs. i8g (Attached SSD)
  While i8g instances (storage-optimized ARM with attached SSDs) showed excellent cold query performance, the m8g series with EBS emerged as our strategic choice:
• Operational Excellence with EBS: The ability to dynamically resize disks to perfectly fit your needs, and enjoy smoother generation upgrades without being tied to fixed, often inconsistent attached SSD sizes is invaluable for a managed service and for our users.
• Competitive Hot Query Performance: For many common workloads (where data is cached), m8g performance was very close to i8g.
• Better Scalability & Cost at Larger Sizes: EBS allows for larger instances with increased CPU/RAM without the need for additional, often unnecessary, attached storage. This flexibility helps reduce overall costs by decoupling compute from excessive storage.
• More CPU Threads: m8g instances generally offer more vCPUs at comparable price points or memory configurations than i*g, aiding CPU-intensive background tasks like ClickHouse’s merges and supporting more concurrent client connections, very valuable for intensive usage of Aiven for ClickHouse.
• Fresher & More Available: The m series tends to be updated more regularly by AWS and has broader regional availability. This helps us keep your service on the cutting edge. The i3en (our old instance) came out in 2019, and its like-for-like SSD size successor, the i7ie (Intel, not even ARM!), only arrived in 2024 and is much more expensive.

WINNER AWS: m8g instances with AWS EBS gp3.

For GCP: C4 family with Hyperdisk

The GCP landscape was more nuanced, with less dramatic differences between families, especially N-series and C-series instances:

• Balanced Performance: Both N-series (general-purpose) and C-series (compute-optimized) instances showed close performance.
• Hyperdisk Advantage: We opted for the flexible Hyperdisk over attached fixed SSDs, to offer the possibility of optimizing Aiven clusters for storage more granularly.
• Standardized Memory: Moving to standard memory configurations (4GB RAM/vCPU) for new N4/C4 instances provides a better CPU-to-RAM balance for most ClickHouse workloads, reducing costs by avoiding overspending for unused memory. For example, C3 instances with SSDs were, in some cases, slightly faster than C4/N4. However, the difference wasn't substantial, and C3 instances have small, fixed SSDs and inconsistent CPU numbers. This configuration does not align with the flexibility we aim to provide our customers.

WINNERS GCP: c4-standard** and especially their ARM equivalent** c4a-standard with GCP Hyperdisk with the best price/performance ratio overall.

How this translates to a better Aiven experience for you

In a nutshell:

1. Superior Price-Performance: Get more analytical power for your money! Depending on your chosen region and plan, you can now double your compute power for the same price.
2. Dynamic Disk Flexibility: Thanks to AWS EBS and GCP Hyperdisk, you have greater control over your storage, allowing for expansion without downtime and freedom from fixed-size SSDs.
3. Future-Proofed Platform: Opting for trusted, frequently updated instance families - particularly ARM on AWS - ensures your service leverages continuous cloud provider advancements. This transition from older, less efficient instances to the newest generations optimizes for both cost and speed.
4. Optimized for Real Workloads: Our choices are based on rigorous, ClickHouse-specific benchmarks, not just paper specs.
5. Peace of Mind: We’ve done the complex VM evaluation so you don’t have to. You can focus on your data and business logic, confident that the underlying infrastructure is top-notch.

Conclusion: Our Commitment to Your ClickHouse Success

Choosing Aiven for ClickHouse means you get the benefit of cutting-edge hardware, flexible storage, and the peace of mind knowing that we're constantly striving to provide the best possible platform for your analytical needs, ensuring optimal value for your investment.

We're excited to roll out these new plans! Start a new service with those new plans today.
For existing Aiven for ClickHouse services, migrations will be as smooth as a regular plan upgrade, directly in the console or with the Aiven API. Check out our FAQs highlighting all the processes to get started with a new plan.
Got questions about migrating or choosing the right plan? Contact our ClickHouse experts right now.