Both Arcfra and Nutanix provide cloud platforms for enterprise users to modernize IT infrastructure. What’s the difference between Arcfra Enterprise Cloud Platform (AECP) and Nutanix Cloud Platform? What features and capabilities do they offer? How do their core components (like virtualization and storage) work? How well do they perform in application environments? This article will provide exact answers to these questions.

Products and Key Capabilities

Arcfra Enterprise Cloud Platform

Nutanix Cloud Platform

In terms of cloud platform components, both Arcfra and Nutanix offer full-stack infrastructure including computing, storage (distributed block and file), management, network & security (distributed firewall, LB, VPC, network visualization), Kubernetes engine, data protection, Active-Active, etc. AECP also supports mainstream servers including Nutanix OEM appliances, which allows users to continue using existing servers and save hardware investment.

Comparison of Key Component: Virtualization

Both Arcfra and Nutanix offer native hypervisors — Arcfra Virtualization Engine (AVE) and AHV — which are developed based on KVM. Therefore, AVE and AHV show similar performance while AVE provides more enhanced features tailored for enterprise users.

Key differences between AVE and AHV in features

Comparison of Key Component: Storage

Currently, distributed storage architecture is divided into two categories: those based on Ceph, an open-source technology, and those based on independently developed technologies. Nutanix and Arcfra both have independently developed storage architectures based on GFS (Google File System). However, they may differ in certain storage architecture designs, including metadata service, data storage engine, and consistency protocol.

Cluster Core Component Relationships

ABS (left) vs. AOS (right)

While the two solutions use similar components for various services, ABS has implemented technological enhancement for some core components such as the metadata service. This is because both Zookeeper and Cassandra have some shortages. On the one hand, Zookeeper has very limited data storage capacity and cannot be used in conjunction with data services. Cassandra, on the other hand, lacks an ACID mechanism, which increases the complexity of upper-layer implementation and thus requires additional effort. ABS, however, combines LevelDB and Zookeeper, making this service more reliable, high-performance, and lightweight.

Storage Data Structure

ABS (left) vs. NDFS (right)

The two solutions adopt different block unit sizes. ABS uses a data unit (extent) of a larger granularity, which can improve access efficiency by reducing the memory resources consumed by metadata and ensuring that all metadata can be stored in memory. Notably, ABS also supports thick provisioning.

I/O Path

ABS (left) vs. NDFS (right)

Both solutions share similar features concerning the I/O path such as automatic tiering of data, data locality, and automatic data balancing. However, ABS can be more efficient as its metadata is cached in memory. This speeds up the response of operations concerning metadata. Moreover, ABS divides its cache into two layers (write and read), which results in different read I/O paths when using replica and EC strategies. Notably, ABS can provide high read/write performance whichever the data redundancy strategy is used.

Data Redundancy and Reliability Features

Comparison of Snapshot Mechanisms

Both solutions use similar technologies for data redundancy and reliability. However, ABS’s replica allocation policies are more diversified. For example, it enables localization and dynamic adjustment of replicas (based on the storage capacity). Also, as the metadata of ABS snapshots is stored in the metadata service cluster, it can be more responsive and enables data to be persistently synchronized to SSD media. Even if a host is rebooted, metadata can be quickly loaded into memory via SSD without causing snapshot performance degradation.

Performance

Different storage designs result in variations in storage performance. We conducted the benchmark performance test (using FIO) and Oracle performance test on Arcfra’s AECP and Nutanix HCI (AOS 6.5). The following examinations were based on the same physical devices and environments. AECP was tested in one volume while Nutanix AOS was tested in one or multiple volumes according to its best practice.

Test Environment

Benchmark Performance Tests

In the benchmark test, AECP outperformed Nutanix AOS except for the sequential read and mixed-read-write tests.

Oracle Performance Tests

Data Import & Export

In database import and export scenarios, we used the SOE (Simple Order Entry) model to measure the time required for importing and exporting data of different volumes, with shorter times indicating better performance. Overall, the tests showed that AECP’s single-volume performance is significantly superior to Nutanix HCI’s single-volume and comparable to Nutanix-ASM (best-practice configuration). Additionally, in data export tests, AECP demonstrated markedly better performance than both Nutanix’s single volume and its best-practice configuration.

TPC-C (Transaction Processing Performance Council)

In the TPC-C test, we used Swingbench for stress testing, where higher TPS (Transactions Per Second) indicates better performance. The results showed that as concurrency increased, the average OPS (Operations Per Second) for both solutions rose significantly. Notably, the single-volume performance of AECP is markedly higher than Nutanix’s single-volume deployment and slightly superior to Nutanix-ASM’s performance.

Securities Order Insertion

The test was based on a stress testing script provided by a counter trading software, focusing on insert operations for the order table to simulate the scenario of orders being written to the database. Each script is committed after every 1,000 transactions and executed in a loop, with a total of 30 PDC scripts running simultaneously for 2 minutes. The test used COUNT(*) to measure the amount of data inserted per second, with higher values indicating better performance. From the results, it can be observed that after 80 seconds, Nutanix AOS experienced multiple instances of performance nearly bottoming out, whereas AECP demonstrated more stable and consistent performance throughout the test.

We further extracted the raw data and excluded the 10 highest and 10 lowest values. The average performance for both systems was then calculated. The overall single-volume performance of AECP surpasses that of Nutanix single-volume deployment (with a performance improvement of over 60%) and the best-practice configuration (with a performance improvement of approximately 5%).

Conclusion

Given the technical optimizations of ABS, AECP delivers superior performance in database scenarios; AECP achieves optimal performance with a single-volume deployment, whereas Nutanix HCI requires multi-volume configurations and relies on system or application-level parallel scheduling to provide comparable performance.

Migrate from Nutanix to Arcfra

Replacing Nutanix with Arcfra can be a breeze! Arcfra offers four solutions for migrating virtual machines from Nutanix to AECP clusters, tailored to different deployment environments and replacement objectives to meet various user needs:

1. Replace Nutanix AOS while retaining VMware virtualization.
2. Replace both VMware virtualization and Nutanix AOS.
3. Replace both Nutanix AHV and AOS.
4. Replace Hyper-V/XenServer virtualization and Nutanix AOS.

To learn more about AECP, please visit our website.

Reference:

1. The Nutanix Cloud Bible https://www.nutanixbible.com/4c-book-of-aos-storage.html

Notice: The feature and performance comparisons between Arcfra and Nutanix were conducted based on AECP preview version and Nutanix AOS 6.5 version in the lab environment. Please refer to Nutanix’s technical documents for its official product features and performance demonstration.