How to Measure Server Efficiency with SERT
SERT™ stands for Server Efficiency Rating Tool and it was developed by SPEC® in collaboration with the U.S. Environmental Protection Agency. It collects measurements and uses an accompanying efficiency metric to assess computer server energy efficiency for ENERGY STAR certification. The tool is now used by the ENERGY STAR program and the EU Lot 9 EcoDesign regulation, as well as Japan’s Top Runner program for computer servers, as a harmonized way to evaluate computer server efficiency data in different geographies across the globe.
SERT is organized around eleven worklets which broadly fall under three categories: CPU based, memory based, and storage based. The tested results from the worklets are aggregated into a single score with a weighting of 65% for CPU, 30% for memory, and 5% for storage worklets. SERT measures the power demand of these different worklet groupings at idle as well as several designated utilization levels to capture variations in workflow. These measurements allow users to make better informed decisions about which computer servers provide the best performance and energy balance for their specific workloads, power management settings, and deployment size. More details on how to use SERT to make better purchasing decisions can be found here.
A Peek Under the Hood of the SERT Score and Worklets
The Green Grid’s SERT Overview goes into great detail regarding how the SERT score is compiled. A hierarchy of how the different worklets and utilizations roll up into the final efficiency score for a computer server is shown below:
In the chart above, the percentages represent utilization levels that the server is measured at for each of the worklets in the test. This leads to an obvious question, what exactly are all these synthetic worklets and what do they represent in the real world? The table below summarizes each worklet and what task is being performed during the test.
|
Worklet
|
Calculation / Operation Type
|
Worklet Description
|
|
Compress |
Compression |
Compresses and decompresses data using a modified Lempel-Ziv-Welch (LZW) method |
|
CryptoAES |
Encryption |
Encrypts and decrypts data using the AES (or DES) block cipher algorithms |
|
LU |
Floating Point Calculation |
Computers the LU factorization of a dense matrix using partial pivoting. Exercises linear algebra kernels (BLAS) and dense matrix operations |
|
SHA256 |
Encryption |
Utilizes standard Java functions to perform SHA-256 transformations on a byte array |
|
SOR |
Floating Point Calculation |
The Jacobi Successive Over-relaxation (SOR) exercises typical access patterns in finite difference applications such as solving Laplace's equation in 2D with Drichlet boundary conditions |
|
SORT |
Sorting |
Sorts a randomized 64-bit integer array during each transaction using java.util.Arrays.sort ( ) API |
|
SSJ |
Simulated Online Transaction Processing |
Exercises the CPU(s), caches and memory of the UUT. Executes six transactions; New order (30.3%), Payment (30.3%), Order Status (3.0%), Delivery (3.0%), Stock Level (3.0%) and Customer Report (30.3%) |
|
Flood 3 |
Memory Bandwidth |
Sequential memory bandwidth test that exercises memory using arithmetic operations and copy instructions. Flood is multi-threaded to reward servers that can utilize more memory concurrently with multiple CPUs and DIMMs |
|
Capacity 3 |
XML Validation |
Performs Java's XML operations on a minimum and maximum data set, using the largest data set that the system can contain in memory with minimal performance penalties. |
|
Random |
Read/Write |
Reads and writes data to and from random file locations |
|
Sequential |
Read/Write |
Reads and writes data to and from file locations that are picked sequentially |
|
Idle |
None |
State of readiness to execute a workload, but no workload is present |
The efficiency score, which results from aggregating the normalized scores of the worklet tests above, can be used by data center operators to understand which computer servers are best suited from a work/watt standpoint to serve their needs. The tool also allows users to create baseline operation cases and then tweak power management and/or performance settings to fine tune their computer servers to find the optimized balance of performance and energy use in their data center.
Many SERT measurements are available on the ENERGY STAR certified product list, so head over there and take a look at how you can get the most efficient server to fit your needs. Are you a manufacturer? You can license SERT to directly measure your specific data center configurations and adjust as needed. Looking for more information? Explore our guides on idle power in your server and how to best utilize SERT to purchase the right ENERGY STAR certified server for your installation.
Author: Ryan Fogle
