BMD best practice for virtualization and hardware design

The topic of virtualization has evolved steadily over the past few years and is indispensable in today's world of IT infrastructure. Below you will find a few useful tips and tricks on how to get the most out of BMD software in your virtual environment. As a reference we use a current HP ProLiant 9th generation server with Intel Xeon E5 v3 processor technology and DDR4 RAM under VMWare ESXi 6.0.

Recommended CPU:

BMD software is a single-core application, which means that it benefits most from a processor with a higher clock frequency rather than from a processor with many cores. The use of so-called entry-level and low-voltage CPUs is therefore not advised. Modern processors also have various energy-saving mechanisms. However, they significantly limit performance, so these power options should always be configured for maximum performance. 

Recommended RAM: 

As for RAM, there are no special requirements regarding the BMD software. However, experience has shown that an even distribution to all memory channels and the use of dual-rank modules offer the best performance. Any "overcommitment" of the RAM should be avoided. 

Recommended drives: 

Drives are supposedly the most important factor for optimal operation. The more virtual machines need to share the available memory, the faster it should be. The use of SSDs contributes to a significant increase in performance. However, there are some additional aspects that are important. 

When using SAS hard disks: 

• Choice of the array: As a rule of thumb, use RAID 10 for optimal performance
• Number of plates: As a rule of thumb, the more plates the faster 
• When choosing between 10k and 15k plates: The extra power usually does not justify the price difference. It is advised to invest in more 10k boards rather than in expensive 15k boards. 

When using SSDs: 

When used for servers, SSDs are generally divided into different price/performance categories: 

• Read Intensive
• Mixed Use
• Write Intensive 

If you want to use an SSD as both a read and write cache, only the models of the categories "mixed use" and "write intensive" are recommended. 

In the test scenario below, we compare RAID5, RAID50 and RAID10 in combination with an SSD as a cache to their counterparts. The only aspect measured, was random access (random I/O). The transfer rate (sequential I/O) is not valid enough for a practical comparison and is therefore not listed. 

As for reading, the matter is clear: There is a major increase in the performance due to the SSD. Furthermore, it is apparent that performance is also related to the different RAID levels as well as the number of hard disks (RAID5 > RAID50 > RAID 10). 

When it comes to the write access, we have to take a closer look. Although the SSD does not respond as quickly as the array controller's cache itself, performance does profit noticeably from the higher cache size of the SSD. The difference in performance between the various configurations is even more visible when writing rather than reading: RAID5 is the slowest, RAID50 offers a good balance between performance and capacity, and RAID10 is always the fastest. 

Conclusion: 

We do not necessarily agree with the statement that an installation in a virtual environment is always slower than a physical installation. 

Based on our experience, however, it can be said that with an appropriately dimnesioned hardware only very little power is lost. 

In view of the many advantages of VMWare, Hyper-V and similar hypervisors, we can definitely recommend the use of a bare-metal virtualization solution.