Skip to main content

Azure Table Performance - 1 vs 100.000 Tables under the same Storage Account

In our system we are using Azure Table to store a list of commands that needs to be send our clients and persisted until the client is available. Because the number of clients is high (more than 100.000), it would be very expensive to store the list of commands in other resources like Redis Cache or SQL Azure.
From the performance perspective, Azure Table are amazing, very fast even at high throughput when you store a lot of data inside them.

At the first version we done a simple mapping, where we had only one Azure Table for all our clients. For each client, we had a dedicated partition in the table. This works great because Azure Table is partitioned (scale) based on the partition.


There is only a small problem with this approach and is related to maintenance and support. If a support engineering needs to look at the commands of a specific user it will be hard for him to navigate and access the data.

The second approach is to create a different Azure Table for each client. The current documentation specify that we can have as many tables we want under a Storage Account without affecting the performance.


Before doing such a change in our system we decided to run a performance test and see if the performance is impacted in one way or another if we have on one table that is big of 100.000 tables.

We run 3 different scenarios with the same load on Azure Table:
  • One big table with all the commands inside it
  • 100.000 empty tables (one per client), were clients only checked if they have commands
  • 1000.000 tables (one per client), that had 5 commands for each client
The source of the load were on-premises machine. Don't focus on the base latency, but the different between this 3 different scenarios. When we access Azure Table from Azure environment (like Worker Roles), the latency for a read operation is under 10ms.

Results are express in milliseconds and is the average of multiple runs.



As we can see there is no impact having 100.000 tables under Azure Storage or one. Based on your needs, it might be more simple to have multiple tables, especially when you need to be able to run execute cleanup steps on large amounts on data . Accessing tables partitions and delete row by row will be expensive and time consumption. Deleting a whole Azure Table can be done with only one simple request.
We can even say, based on current results that you have better performance if you use multiple tables and not only one.

Comments

Popular posts from this blog

How to check in AngularJS if a service was register or not

There are cases when you need to check in a service or a controller was register in AngularJS.
For example a valid use case is when you have the same implementation running on multiple application. In this case, you may want to intercept the HTTP provider and add a custom step there. This step don’t needs to run on all the application, only in the one where the service exist and register.
A solution for this case would be to have a flag in the configuration that specify this. In the core you would have an IF that would check the value of this flag.
Another solution is to check if a specific service was register in AngularJS or not. If the service was register that you would execute your own logic.
To check if a service was register or not in AngularJS container you need to call the ‘has’ method of ‘inhector’. It will return TRUE if the service was register.
if ($injector.has('httpInterceptorService')) { $httpProvider.interceptors.push('httpInterceptorService&#…

ADO.NET provider with invariant name 'System.Data.SqlClient' could not be loaded

Today blog post will be started with the following error when running DB tests on the CI machine:
threw exception: System.InvalidOperationException: The Entity Framework provider type 'System.Data.Entity.SqlServer.SqlProviderServices, EntityFramework.SqlServer' registered in the application config file for the ADO.NET provider with invariant name 'System.Data.SqlClient' could not be loaded. Make sure that the assembly-qualified name is used and that the assembly is available to the running application. See http://go.microsoft.com/fwlink/?LinkId=260882 for more information. at System.Data.Entity.Infrastructure.DependencyResolution.ProviderServicesFactory.GetInstance(String providerTypeName, String providerInvariantName) This error happened only on the Continuous Integration machine. On the devs machines, everything has fine. The classic problem – on my machine it’s working. The CI has the following configuration:

TeamCity.NET 4.51EF 6.0.2VS2013
It seems that there …

[Post-Event] Codecamp Conference Cluj-Napoca - Nov 19, 2016

Last day I was invited to another Codecamp Conference, that took place in Cluj-Napoca. Like other Codecamp Conferences, the event was very big, with more than 1.000 participants and 70 sessions. There were 10 tracks in parallel, so it was pretty hard to decide at  what session you want to join.
It was great to join this conference and I hope that you discovered something new during the conference.
At this event I talked about Azure IoT Hub and how we can use it to connect devices from the field. I had a lot of demos using Raspberry PI 3 and Simplelink SensorTag. Most of the samples were written in C++ and Node.JS and people were impressed that even if we are using Microsoft technologies, we are not limited to C# and .NET. World and Microsoft are changing so fast. Just looking and Azure IoT Hub and new features that were launched and I'm pressed (Jobs, Methods, Device Twin).
On backend my demos covered Stream Analytics, Event Hub, Azure Object Storage and DocumentDB.

Title:
What abo…