Shulou(Shulou.com)06/01 Report--

Compare the performance of MongoDB on public clouds: AWS, Azure, and Digital Ocean

Original English text:

Http://blog.mongodirector.com/comparing-mongodb-performance-on-public-clouds-aws-azure-digital-ocean/

A common problem we get from MongoDirector.com is the relative performance of MongoDB in various public clouds such as AWS, Azure, Digital Ocean, and so on. Each cloud makes specific choices for disk architectures that seriously affect MongoDB performance.

So before you spend a lot of time and effort on a particular cloud, it's important to understand the overall performance characteristics of MongoDB on that cloud. We looked for this information and didn't find it-so we decided to integrate it for you as part of our performance series.

Reference platform

For this test, we decided to compare AWS, Azure, and Digital Ocean. We chose two different sets of configurations. The following table summarizes the machine configuration:

Provider

Region

MongoDirector Medium*

(Cores/RAM/Disk/Prov IOPS)

MongoDirector Large* (Cores/RAM/Disk/Prov IOPS)

AWS

US East

1/3.75GB/60GB/300

2/7.5GB/120GB/500

Azure

East US

2/3.5GB/60GB/upto 2000

4/7GB/120GB/upto 4000

Digital Ocean

New York 3

2go to 4GB, to 25GB, to SSD, to *

4According to 8GB, 35GB, and SSDbands *

* refer to the details of the machine configuration under "MongoDB Hosting" here.

* * SSD has been directly attached to Digital Ocean.

The performance benchmark was run using YCSB load A (load with a large number of updates). We talked about YCSB, configuring it, and its load in a very detailed article last month.

1. All benchmarks are performed in a single configuration.

two。 For all configurations, insert 5 million records using various levels of server load (based on a large number of client threads).

3. For medium configurations, load A uses various levels of server load, with a total of 5 million operations, at the default value (50% update, 50% read).

4. For large configurations, load A uses various levels of server load, with a total of 10 million operations, at the default value (50% update, 50% read).

Result

We will discuss the results based on insertion performance and throughput / latency characteristics under a large update load.

Insertion performance

Medium instance

In a medium configuration, for the throughput / latency (Throughput/latency) characteristic of inserting 5m records:

Large scale instance

In large configurations, insert throughput / latency (Throughput/latency) characteristics of 5m records:

Update performanc

Medium instance

"in a medium configuration, for write / update 5m operation throughput / latency (Throughput/latency) characteristics:"

This test is only run with 32 threads for Digital Ocean. AWS and Azure are horizontal at 16 threads. However, Digital Ocean gives the impression that it does not grow linearly until 32 threads.

Large scale instance

"in large configurations, throughput / latency (Throughput/latency) characteristics for write / update 10m operations:"

Whole analysis

1. As expected, Digital Ocean has always had a consistent high throughput / low latency feature and beat its competitors in the plug-in phase to get the maximum performance from its native SSD driver. Interestingly, even if it works well during the read / update phase, other providers give a little bit of competition, especially when the server load increases. It is clear that AWS/Azure uses network storage with higher throughput.

two。 To get better performance from AWS disks, users can use larger disk sizes or allocate higher-precision IOPS.

3. In medium instances, Azure always seems to do better than AWS in the insert and update / read phases. This is a pleasant surprise. The hardware is completely equal. In large instances, AWS performs significantly better than Azure.

4. The latency of AWS and Azure decreases well with the increase of load. Azure seems to have a better delay reduction curve.

5. Another interesting aspect of AWS performance is how flat it always is: it is elegantly reduced even on the logarithmic scale.

6. Based on the amount of latency, from a load standpoint, there are 8 and 16 threads for medium and large instances, respectively, which looks like hot spots.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.