Earlier this spring, we migrated the infrastructure to the updated Xeon Gold 6254. It's too late to provide a detailed review of the processor — more than a year has passed since its release, and everyone is familiar with its details. However, one notable feature is that the processor has a base frequency of 3.1 GHz and 18 cores, which can all operate simultaneously at a frequency of 3.9 GHz under turbo boost, allowing us, as a cloud provider, to consistently deliver high CPU frequency in virtual machines.
Nevertheless, we are still interested in assessing its capabilities under load. Let's get started!
Brief description of the processor
As we mentioned earlier, the processor is widely known, but here are its specifications in brief:
Codename
Cascade Lake
Number of cores
18
Base clock frequency
3.1 GHz
Maximum clock frequency with Turbo Boost technology across all cores
3.9 GHz
Memory types
DDR4-2933
Max number of memory channels
6
Conducting tests
For the test, we prepared a virtual server with 8 dedicated cores and 64 GB of memory, with data located on a fast pool based on an SSD array. Testing is conducted on the Microsoft SQL Server 2014 database, with the operating system being Windows Server 2016, and of course, we can't do without the most important software — 1C:Enterprise 8.3 (8.3.13.1644).
We also took into account . If you haven’t read it, briefly: they tested four processors — 2690, 6244, and 6254. The fastest was the 6244, while the 6254 achieved a score of 27.62. This experience piqued our interest because in earlier tests in our cloud during spring 2020, we were getting Gilev test scores ranging from 33 to 45, but we never went below 30. This might be a feature of working with another DBMS, but it motivated us to conduct our own measurements on our infrastructure. We've done this again and will share our findings.
So, let’s start the testing! What are the results?
Testing results
As we can see, on the MSSQL server with the Xeon Gold 6254 processor and turbo boost enabled, the result is 39 points. We interpret the obtained value into a rating according to Gilev and find the result to be higher than the rating of 'Good', but not quite 'Excellent'. We consider the result to be good from the perspective of assessing this type of 'parrots'. It is important to note that we did not optimize at the OS and SQL server level and received the result as it stands; with some effort, it can be increased slightly, but that pertains to tuning details, which is a topic for a separate blog post.
It should also be noted that we do not encourage assessing the actual workload of production databases based on the Gilev test and making immediate conclusions about the feasibility of using a particular processor. However, according to our statistics, processors with a clock speed of 3 GHz and above will be more effective when working with 1C. The Gilev test can yield different results, even with the same provider or in a local infrastructure. You can achieve high results on simpler processors, even non-server ones, but that does not mean that when you 'feed' a workload in the form of 1C ERP for 50-100 users or Trade, you will get consistently high results. Always conduct a pilot and perform testing if possible.
Source: habr.com
