High Performance MongoDB Clusters on Azure

ScaleGrid now supports 'High performance' MongoDB clusters on the Azure ARM platformClick To Tweet  You can bring your own Azure account or host with us.

High-performance clusters use powerful CPU’s and premium disks to deliver enhanced performance. Based on our performance tests (details below) our High-performance clusters on Azure deliver 4x the performance of our competition and 2x the performance of our standard clusters.

Specifications

High-performance clusters are offered in four configurations as detailed below. You can also see the details on our pricing page. If you are using the management plan the sizes can also be customized.
High performance clusters on Azure

Performance

We ran detailed performance tests using YCSB to demonstrate the improved performance. We compared ScaleGrid High perf XLarge vs ScaleGrid standard XLarge vs Competitor equivalent size.  All configurations are similar with 14GB RAM.

  • Region: Australia East – Sydney
  • ScaleGrid High performance –  HighPerfXLarge (DS3_v2 – 14 GB RAM, 4 cores, 1024 GB, Premium SSD disk, 5000 IOPS). Replica Set 2 Nodes + 1 Arbiter. MongoDB version 3.4.3.
  • ScaleGrid standard – XLarge (D11_v2 – 14 GB memory, 2 cores, 240 GB, 4 disk RAID, up to 2000 IOPS). Replica Set 2 Nodes + 1 Arbiter. MongoDB version 3.4.3
  • Competitor 1 – Azure Dedicated (2 + 1) Replica Set with 14 GB memory and 240 GB block storage per node. MongoDB version 3.4.4

 Workloads

  • 100 % Write workload (Workload A Load phase): Write 3 million records into the database at varied levels of loading by increasing the number of client threads from 4 going up to 64, doubling every time.
  • 50 % Read/ 50% update workload (Workload A run phase): Run workload for 6 million operations at varying levels of loading by increasing the number of client threads from 4 going up to 64, doubling every time.

Results

100% Write workload
Hiperformance mongo clusters - insert workload

50% Read/50% update workload
Hi performance mongo clusters workload A results

 

Summary

  • With no RAID or premium SSD disks, the competitor’s performance is pretty low. It saturates its block storage at low loads and throughput remains almost flat throughout. CPU saturates completely at 16 threads. Latency growth is almost exponential.
  • With RAID configured, the standard ScaleGrid Azure solution greatly outperforms the competitor as it’s throughput increases almost linearly with load. Latency numbers are also very low. This is expected with a RAID setup and a sequential workload. At their best, our standard Azure solution was more than 3 times better than the competitors!
  • ScaleGrid High Perf deployment with its premium disks outperforms the standard block storage. At it’s best it is 5 times as fast the competitor’s solution and about 1.4 times as fast as our standard solution.

If you need a high-performance cluster on Azure reach out to support@scalegrid.io and we can help you get setup.


Dharshan is the founder of ScaleGrid.io (formerly MongoDirector.com). He is an experienced MongoDB developer and administrator. He can be reached for further comment at @dharshanrg


22 Shares
+12
Tweet
Share
Share20
Pin