This post has been republished via RSS; it originally appeared at: Microsoft Tech Community - Latest Blogs - .
No one knows SQL Server like Microsoft, so it should be no surprise that Azure offers the best overall value for running your SQL Server workloads in the cloud. Today I am thrilled to introduce a new Azure Virtual Machines offering that will bring even more value to your SQL Server workloads running in Azure virtual machine: the new memory optimized VM series Ebdsv5 that offers higher remote storage throughput and IOPS per vCPU than any other VM series in Azure.
Hopefully you have already read our previous blog about getting the best price-performance for your SQL Server workloads, How to get the biggest bang for your buck with SQL Server on Azure VMs. If not, I encourage you to read it so you can learn about the importance of sizing and configuring your Azure VMs for the best performance at the lowest cost. In that blog, we focused on the E64-32ds_v4, a great option for SQL Server workloads as it offers an ideal 1:8 core-to-memory ratio, high remote I/O throughput, and a large amount of local throughput for tempdb and Azure BlobCache. The Edsv4 VM series was our favorite for SQL Server workloads until now.
If you’ve been working with SQL Server for a while, or any relational database system for that matter, you know how critical I/O performance is to the overall performance of your database application. Relational database systems generate a large amount of I/O per core, much higher than most other workloads. The new Ebdsv5 series offers the highest I/O throughput to core ratio available in Azure and is ideal for SQL Server workloads and other I/O-heavy workloads like Oracle, Teradata, and SAS. With the current top end of 4 GBPS remote I/O throughput at 64 vCPUs, these preview sizes will cover most of your SQL Server needs. Once the Ebdsv5 series is generally available we expect it to be the flagship Azure VM for SQL Server workloads.
If you’re already running SQL Server in Azure VM, you may be wondering if it’s worth switching over to this new size once it’s available. The answer to that question is, as with any SQL Server question, it depends. Many customers find that in order to support this need for high I/O throughput with the current generation of VMs, they must leverage constrained core VMs. The E64-32ds_v4 is an example of a constrained core VM. It offers the memory and I/O throughput of E64ds_v4 but with only 32 cores. This means you get more memory and throughput per vCPU, and you only pay SQL Server licensing for 32 cores. With the new E32bds_v5 VM, you can get even more I/O throughput than the E64-32ds_v4 VM at a fraction of the cost. Here’s a table that compares the two VMs:
Check that out! More than DOUBLE the remote I/O throughput at nearly half the cost!
We’ve been doing preliminary testing with these VMs, and while we still have some work to do before they are generally available, the results are already very promising. As we did for the previous blog, we used HammerDB to execute a TPC-C1 style workload against the new VM. The performance metric used for these tests is New Orders Per Minute (NOPM). The following table compares one of our early E32bds_v5 test runs with the runs we did on the E64-32ds_v4 VM for the previous blog:
As you can see, we got nearly the same level of performance at close to 30% lower total cost!
If you’d like to try out one of these VMs for yourself, sign up for the limited public preview at https://aka.ms/signupEbsv5Preview.
1The HammerDB TPC-C workload is derived from the TPC-C Benchmark and is not comparable to published TPC-C Benchmark results, as the HammerDB TPC-C workload results do not fully comply with the TPC-C Benchmark.