 |
|
USB Disks Performance Improvement
The performance of external USB drives depends on each specific drive type, model, the type of the computer, etc. In addition, for a given hardware configuration, there are a number of software controlled options, which can improve the performance of external USB drives. One of these options is the disk write cache, which can be enabled or disabled according to the user-specific needs. 
By default, in order to prevent potential errors with frequently disconnected USB drives, the operating system disables the disk write cache for all external USB drives. But, some external USB drives are permanently connected to the computer and for such configurations the disk write cache may be enabled without any problems. The purpose of this performance review is to show the potential performance impact of the disk write cache for a number of different types of external USB drives using different sets of files. For this review, we have performed a series of performance tests using 3 different types of external USB drives with 3 different sets of files. For each disk, the same set of tests was performed two times: the first time with the disk write cache disabled and the second time with the disk write cache enabled. For this performance review, we have used the following sets of files:
All sets of files were copied multiple times from a Samsung 850 PRO 512 GB SSD disk to the following external USB disk drives connected via USB 3.0:
All tests were performed using SyncBreeze v9.0 (64-Bit version) installed on a quad-core Windows Server 2012 R2 with 16 GB of system memory. For each set of files and each tested storage device, file synchronization operations were configured to copy files using 1, 2, 3 and 4 parallel file synchronization threads, which can be set on the advanced file synchronization options 'Speed' tab. File Copy Operations - Mechanical HDD #1
When copying the set of small files to the first mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 10.46 MB/sec for a single file copy thread and scales to 17.63 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 25.99 MB/sec for a single file copy thread and scales to 31.55 MB/sec for 4 parallel file copy threads, which represents an average performance improvement of 102 percent. 
When copying the set of medium-sized files to the first mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 47.08 MB/sec for a single file copy thread and scales to 53.61 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 57.17 MB/sec for a single file copy thread, scales to 60.09 MB/sec for 3 parallel file copy threads and then slightly declines to 59.19 MB/Sec for 4 parallel file copy threads, which represents an average performance improvement of 15 percent. 
When copying the set of large files to the first mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 95.59 MB/sec for a single file copy thread and then declines to 86.21 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 94.69 MB/sec for a single file copy thread and then declines to 85.58 MB/sec for 4 parallel file copy threads, which represents an average performance decline of 1 percent. 
According to the performance results for this specific external USB disk, the disk write cache provides a significant performance improvement with the set of small files, a slight performance improvement with the set of medium-sized files and no performance improvements with the set of large files. File Copy Operations - Mechanical HDD #2
When copying the set of small files to the second mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 6.4 MB/sec for a single file copy thread and scales to 7.25 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 7.9 MB/sec for a single file copy thread and scales to 9.15 MB/sec for 4 parallel file copy threads, which represents an average performance improvement of 20 percent. 
When copying the set of medium-sized files to the second mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 24.04 MB/sec for a single file copy thread and scales to 25.59 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 26.82 MB/sec for a single file copy thread and scales to 28.11 MB/sec for 4 parallel file copy threads, which represents an average performance improvement of 12.68 percent. 
When copying the set of large files to the second mechanical HDD drive with the disk write cache disabled, the file copy performance reaches 86.21 MB/sec for a single file copy thread and then declines to 68.3 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 99.47 MB/sec for a single file copy thread and then declines to 69.53 MB/sec for 4 parallel file copy threads, which represents an average performance improvement of 7 percent relative to the configuration with the disk write cache disabled. 
According to the performance results for the second mechanical HDD, the disk write cache provides not so significant performance improvement with the set of small files as the first mechanical HDD, a slight performance improvement with the set of medium-sized files and what is more interesting a slight performance improvement with the set of large files, which leads to the conclusion that the potential performance improvement of the disk write cache depends on the type and architecture of each specific external USB disk drive. File Copy Operations - USB Flash Drive
When copying the set of small files to the USB Flash drive with the disk write cache disabled, the file copy performance reaches 7.22 MB/sec for a single file copy thread and slightly increases to 8.08 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 12.47 MB/sec for a single file copy thread and then declines to 10.6 MB/sec for 4 parallel file copy threads, which represents an average performance improvement of 42 percent relative to the configuration with the disk write cache disabled. 
When copying the set of medium-sized files to the USB Flash drive with the disk write cache disabled, the file copy performance reaches 26.67 MB/sec for a single file copy thread and then declines to 19.62 MB/sec for 4 parallel file copy threads. With the disk write cache enabled, the file copy performance reaches 26.62 MB/sec for a single file copy thread and then declines to 17.03 MB/sec for 4 parallel file copy threads, which represents an average performance decline of 10 percent relative to the configuration with the disk write cache disabled. 
When copying the set of large files to the USB Flash drive with the disk write cache disabled, the file copy performance reaches 73.59 MB/sec for a single file copy thread and then declines to 15.59 MB/sec for 4 parallel file copy threads. 
With the disk write cache enabled, the file copy performance reaches 60.17 MB/sec for a single file copy thread and then declines to 16.81 MB/sec for 4 parallel file copy threads, which represents an average performance decline of 1.68 percent relative to the configuration with the disk write cache disabled. 
According to the average performance results, the disk write cache delivers the most significant improvement of 78 percent when copying small files using a single file copy thread, which decreases to 44 percent when copying small files using 4 parallel file copy threads. 
When copying the set of medium-sized files, the disk write cache provides a 12 percent performance improvement with a single file copy thread, which drops to just 2 percent when copying files using 4 parallel file copy threads. 
On the other hand, when copying large files, the disk write cache provides no performance improvements and in some configurations a slight performance decline of 1-2 percent is possible. File Delete Operations
In addition to the file copy operations, the disk write cache can potentially improve the performance of file rename and delete operations, which require a large number of small changes to the file system when renaming or deleting large numbers of files. 
For example, the performance of file delete operations for the first mechanical HDD with the disk write cache disabled reaches 894 Files/Sec when deleting files using a single file delete thread and scales to 1969 Files/Sec when deleting files using 4 parallel threads. On the other hand, with the disk write cache enabled, the performance of the file delete operations reaches 4809 Files/Sec with a single file delete thread and scales to 6443 Files/Sec with 4 parallel file delete threads, which represents a 300 percent improvement relative to the configuration with the disk write cache disabled. 
The performance of file delete operations for the second mechanical HDD with the disk write cache disabled reaches 633 Files/Sec when deleting files using a single file delete thread and scales to 1255 Files/Sec when deleting files using 4 parallel threads. With the disk write cache enabled, the performance of the file delete operations reaches 3781 Files/Sec with a single file delete thread, scales to 3866 Files/Sec with 2 parallel file delete threads and then slightly declines to 3701 Files/Sec with 4 parallel file delete threads. 
For the USB Flash drive, the performance of file delete operations with the disk write cache disabled reaches 711 Files/Sec when deleting files using a single file delete thread and scales to 1195 Files/Sec when deleting files using 4 parallel threads. With the disk write cache enabled, the performance of the file delete operations reaches 4065 Files/Sec with a single file delete thread and scales to 4843 Files/Sec with 4 parallel file delete threads, which represents almost 400 percents of improvement relative to the configuration with the disk write cache disabled. 
Averaged for all tested disk drives, the performance improvement for file delete operations reached 386 percents with a single file delete thread and then decreased to 228 percent with 4 parallel file delete threads, which definitely represents a huge improvement relative to the configuration with the disk write cache disabled.
|
   |
