– Run analytical workloads for one-fourth of the AWS costs

Photo by Kevin Ku: Pexels

Cloud computing is the on-demand delivery of IT resources over the Internet with pay-as-you-go pricing. Instead of buying, owning, and maintaining physical data centers and servers, you can access technology services, such as computing power, storage, and databases, on an as-needed basis from a cloud provider.. – AWS

Bare metal servers are a form of cloud service in which the user rents a physical machine from a provider that is not shared with any other tenants. – IBM

According to AWS, the benefits of cloud computing over on-premise computing are

  • Pay-as-you-use model and don’t have to invest in data centers
  • Benefit from massive economies of scale
  • Stop spending money running and maintaining data centers
  • Benefit from OpEx spending instead of CapEx spending

While these are true with any cloud provider such as AWS, Azure, and Google Cloud, it comes with a cost. While these options may seem beneficial and cheaper in the short run, cloud spending can cost a fortune in the long run.

Bare metal servers offer the benefits of cloud computing which were listed above in this article. On top of that, it offers significant cost benefits over cloud compute resources like AWS EC2, Azure VMs, etc. Bare-metal servers also offer the benefits of on-prem and self-managed servers. So the bare metal servers offer the best of both worlds of cloud and on-prem computing. is a bare-metal server provider powered by world-class data centers. They offer high-quality servers at lower prices and come with fixed-price billing. The main focus of is on providing large, high-performing servers to meet the demand for the explosion of data and the need for real-time analytics/detection for this data.

In this article, we will compare the performance of’s server with AWS EC2 and compare the performance of these two hardware on analytical workloads using ClickHouse.

ClickHouse Benchmark

ClickHouse is a popular open-source database management system for OLAP workloads and real-time analytics. Notably, ClickHouse is optimized for speed and offers great performance at a significantly lower cost when compared to its competition. 

ClickBench, a benchmark for analytical databases, uses a standard dataset and measures the execution time for the pre-defined queries. ClickBench is one of the most used analytical database benchmarks in recent times. You can find the results for prominent databases here. The dataset consists of 99,997,497 records which is approximately 75 GB. 43 queries are run thrice (1 Cold + 2 Hot runs) as a standard practice.

This report is to compare the performance of ClickHouse on ClickBench which was run on an AWS EC2 instance and a bare metal server powered by (powered by BitRefinery). 

Pricing Comparison

In this comparative study, we will run the ClickBench on a bare-metal server (Gold Server) and use the readily available results for the AWS EC2 instance (c6a.metal instance at US-East-Ohio).

Let us look at the hardware and the estimated annual cost of these servers. We have extrapolated the storage and egress costs for AWS to match with  Bare-Metal hardware specs.


Specification Bare-Metal AWS EC2 (c6a.metal)
CPU 80 Core 192 vCPU
RAM (GB) 1024  384
Storage (TB) 40 (Raid 6) 40 (GP3)
Network Bandwidth (Intra-region & Internet) 10 Gbps & 1Gbps 50 Gbps & 25 Gbps
Approx Egress via Internet (per Month) 40 TB 40 TB
Annual Cost (USD) $ 36,000 $ 141,792 (dedicated,

1 year reserved)


TLDR: The bare metal server costs roughly one-fourth of the AWS EC2 instance for similar storage and RAM. The AWS gp2 storage has an IOPS limit (3000 IOPS for < 1 TB and up to 16,000 IOPS for up to 16 TB. So increasing the IOPS could increase the costs further.

Comparison Gold Server

  • Cost-effective
  • Single tenant
  • Benefits of cloud as well as on-prem
  • RAID-6 Storage
  • No storage IOPS Limit

AWS EC2 – C6a.metal

  • Auto Scaling
  • Pay-as-you-use billing
  • Better network Bandwidth
  • Highly elastic
  • Data centers spread across the globe

Benchmark Comparison

Query Performance

Both servers were able to run all the queries and the average durations were quite similar for both servers. However, the average query execution durations for some of the queries were taking considerably longer in the c6a.metal instance.

Next, let us compare the average query execution duration for both servers and see which one performs better in the majority of the queries.

Bare Metal server performs better in 58% of the queries and AWS c6a.metal performs better in ~ 42% of the queries that were part of the benchmark.

Cold Run

When the queries are run for the first time, then there would be no disk cache or query cache to speed up the execution. So the Cold run performance can be an indicator of the raw power of the server.

We will compare the mean query execution duration and the median query execution duration for both servers. The difference between the mean query execution duration for bare-metal and AWS c6a.metal servers is quite pronounced. This is because, for some of the queries, the execution duration for c6a.metal is quite high compared to the bare-metal server.

So, let us look at the median values, which are immune to such outliers. The median query execution duration for bare-metal is still lower than the AWS c6a.metal but the difference is lesser than what we have seen while comparing the mean query execution duration.

Hot Run

When we look at the mean and median query execution duration for the hot runs, the AWS c6a.metal is lower than the bare-metal server.


The summary of this comparative study is below.

  1. The Bare-Metal server offers better performance at one-fourth the price as compared to the AWS c6a.metal server
  2. The Bare-Metal server can do this despite less than half the CPU cores – which could mean even better performance on the same number of cores
  3. The c6a.metal server has just 37.5 % of the RAM in the Bare metal server
  4. Bare-Metal does well in cold runs, and AWS c6a.metal has the upper hand in hot runs, even though the relative performance in Hot runs is not too different. 
  5. Overall, the Bare metal server has the upper hand at a considerably lower cost.
  6. With ChistaDATA’s build and performance optimizing methods, we believe that the Price: Performance can be even better for clients.