Thecus N7700PRO Network Storage ServerAug 22nd, 2010 | By Anthony
Hardware specifications are as follows:
How a network attached device performs is dependent on numerous systems both hardware and software. Consequently representing a network attached storage device’s performance is a bit tricky. We use a number of pieces of software in our testing to best provide a picture of what users can and should expect from a product. The difficulty is data transfer is a bit more difficult than sending a few bytes down a tube. It’s a whole lot more involved and when we throw networking in the equation even more so. Perhaps the greatest issue at hand though is theoretical and real world testing.
How a device is accessed and the composition of the data that flows through it is vastly different between instances, clients accessing the device and also especially between types applications. For instance: a network drive being used as a localized storage system for a firm who deals with text documents and emails is quite different from one who deals with digital video. We will use two pieces of software allow us to have a look at real world performance: Intel’s NAS Performance Toolkit and an ex- Intel project, now developed by the OSDL (Open Source Development Lab): Iometer.
Intel’s NASPT offers a number of strengths. First and foremost: consistency. With NASPT’s built in traces, performance numbers resulting in various tests are easily emulated across various systems and gives insight into how a system would perform under real world conditions. Unlike with our more elementary hardware level tests, we are not interested in separating software environmental factors. Day to day storage usage is heavily dependent on system software, operating systems and so on. With the trace files, we can simulate hard drive and network activity so that each and every test is performed under the same conditions.
Iometer is our middle ground test between entirely synthetic theoretical performance and real world. Intel’s NASPT is excellent for providing digested, easy to understand numbers which correspond to applicable situations; however Iometer is more interested in breaking that data down even further. With IOMeter, we control the distribution of random and sequential reads/ writes, the spread of data over sectors- also known as locality, and a number of access characteristics.
The short comings of these two tests are we still don’t have a picture of how things function beneath the surface, how the system deals with different sizes of files and how it deals with caching.
If you’ve been with us here a few times, you would have doubtfully noticed that around these parts we are big fans of IOzone. Unlike NASPT or Iometer, IOzone functions more on the hardware level and does not account for operating system file or network transfer optimization.
IOzone is a benchmarking tool useful for analyzing file system performance on a number of different platforms, including Linux, HP-UX, Solaris, and many others. It uses file system IO as its primary load generator generation, presenting the systems under test with a large range of file IO requests, running from small to very large file sizes, with varying record sizes, while executing performing a number of different file access patterns. It also permits the benchmark administrator to modify system test parameters, affecting the underlying system response to these differing access sizes and patterns in order to illustrate system issues that might not otherwise be obvious, or which might vary from platform to platform.
IOzone is a command line operated testing utility; we will be using the following command.
It should be noted that for the most part, what we are most interested in is the 64kb record size which is most commonly used and is the record size in which we average throughput rates.
Finally, the last piece two pieces of software we will use is ATTO Disk Benchmark and Window’s Robust File copy or Robocopy . For those who just need to know what to expect with all the jargon aside, but need a bit more depth than just skipping to the conclusion, this is for you.
ATTO Disk Benchmark is an easy to test, easy to read and easy to understand benchmarking software. It simply tests read and write throughput over a range of files.
Robocopy in a sense is very similar to NASPT, only less sophisticated and less controlled. Robocopy is a command line representation of copy and pasting files within Windows. Though, as its name suggests it can be quite a bit more robust in terms of running sets of arguments to automatically selectively move only certain files. For our purposes though, it’s handy simply because it outputs an elapsed time and average throughput.
We’ll test by moving two rather large sets of files. The first is a 60GB directory of music which is conventionally organized into the artist>album format along with corresponding album art representing transfers that large in file size but also semi non-sequential in composition. The second is to represent sequential file transfer and is composed of a 15GB directory of videos where each is approximately 1.2GB in size.
Using a battery of both user level and system level tests we can better and more accurately gauge performance and get a better idea of what users can expect.