Comprehensive training on hard drive performance testing; Tips you should know
In order to test the hard drive and check the software and hardware power of HDD and SDD, different methods can be used. In this article, we try to analyze the hard drive test in a convenient way.
The data in this article is based on the hard drive test of companies such as Intel, Crucial, SanDisk, Samsung and Toshiba. . We have looked at the performance of hard drives from several aspects such as packaging, peripheral equipment and product specifications.Buying an external hard drive from SafirSoft
Companies that review hard drives, their data can be It is not generalization to the whole of a product model; Because the type of information obtained is different from one manufacturer to another. On the other hand, even the specifications of different products are not standard. The rule of thumb is based on four issues: sequential read, sequential write, random read, and random write.
Two methods for determining sequential performance are obtained from easy-to-use software with a graphical user interface. Using the ATTO method is an old method to take a deep look at the four mentioned factors. This tool does not allow you to test the hard drive with a single command. SanDisk and several other companies do not use ATTO and instead use CrystalDiskMark to show sequential read and write performance.Buying SSD memory from SafirSoft
Random Input and Output Operations per Second (IOPS) ) can be measured in many ways. Most companies use Iometers with 4 KB blocks at a queue depth of 32. The results are great, but the findings have very little in common with real-world hard drive behavior. When reviewing product specifications on developer websites and product packages, there are a few things to keep in mind. Based on the information we have obtained from the behavior of hard drives in the real world, we suggest that you do not expect to receive relevant data from the hard drive vendor. In general, even normal usage of hard drives comes with varying reports from one day to the next. Since different specifications are used to generate hard drive test data, product information from different hard drive vendor companies should not be compared.
In this article we have tried to generate real and comparable benchmark numbers. However, these numbers and data require a series of specific tests. With solid state drives, whatever workload you run before testing will affect the outcome of whatever you measure next. Naturally, all SSDs run so-called Clean; A condition we call brand new hard drives fresh out of the box (FOB).
FOB status means that the controller is able to write to the flash without the read, change, and write process. Once the drive is filled with data, the drive controller needs to read a page, modify the desired data, and then rewrite the page. This will happen even if the change is only one cell. Read, modify and write processes can double or even triple the latency. This depends on the type of data that has been tampered with.
Hard drive Test Tool
I also test devices that are still in development, often going through multiple firmware edits before release. I check and test SSD. To maintain this procedure and a fair judgment, I need some systems to run things.
I use the above specs and settings to test SSDs and hard drives on four identical systems. These devices are designed to test SATA based products. These systems also test network equipment from time to time. In order for the devices in question not to change, I have prevented the system from accessing the internet to prevent automatic system updates that could affect my results.
PCIe storage spaces are evaluated on a pair of targeted systems. The ASRock Z97 Extreme6 motherboard provides a four-lane PCIe 3.0 connection from the processor to the M.2 interface. This is the most ideal way to connect M.2-based hard drives to a heavy-duty yet high-performance PC. These systems are also disconnected from the Internet. Operating system specifications and test software are kept consistent between the SATA and PCIe testbeds.
I also used several other systems to perform specialized tests, simulating system images of notebook battery life on drives and secure wipe operations. I keep it available. Finally, 29 modern systems, from Sandy Bridge-based notebooks to test hard drive products Dual-Xeon systems are at my disposal for testing home network attached storage (NAS).
We used two different notebooks to test notebook battery life. There are standard SATA 2.5 drives in the Lenovo T440 model; One of the few laptops that supports DEVSLP. I use a Lenovo X1 Carbon Gen 3 notebook to test PCIe and SATA based m.2 SSDs. Lenovo's X1 Carbon Gen 3 notebooks come with an M.2 hard drive. At the time of writing, not many notebooks are available with this feature.
Why is packaging important?
Most of us use online marketplaces to buy products. Sometimes, the availability of a product, apart from considering the packaging, saves money. In any case, wherever you buy the product, packaging is very important.
Online Orders It requires a shipping process and there is nothing worse than waiting to receive a package that is already damaged. In this article, we've also looked at how companies package their hard drives and SSDs. Solid state drives are vibration-proof for the most part. Advances in hard drive technology have increased the amount of vibration that an HDD can withstand. However, we still wish there were vibration-absorbing materials inside the product package.
With SSDs, performance varies with different capacity points. Smaller drives are typically slower than larger models, even within the same family. Some vendors publish specifications for each model, but others only list the maximum power of a series. 128GB and even 256GB models are usually slower than 512GB and 1TB models.
When We shop at a retail store, most of us want to see product information. Again, some drives have the desired information on the box, while other products have very limited information available to the customer. Sellers should not skimp on information.
Four corner test
The four corner hard drive test includes sequential read, sequential write, random read and random write. Not every analyst or company handles these things the same way.
Sequential data is typically measured in 128KB blocks. However, some editors prefer to use blocks of 64 KB and more than 8 MB. For the most part we use 128KB blocks, but we've published a single-threaded chart of block sizes ranging from 512bytes to 8MB for sequential and random workloads. This chart also shows the depth of ranks from 1 to 32.
Random data performance is almost universally benchmarked with 4kB blocks at a depth of 32. This meter and benchmark shows what the manufacturers want users to see, although this information does not represent the actual behavior and performance of the hard drive. We demonstrate the state of random performance with 4 KB blocks at varying depths from 1 to 32 for most devices. PCIe-based products cope well with this depth. Therefore, in some tests we have progressed to 128 KB.
In each review, we show a comparison between sequential reads and writes at the queue depth of both. We also plot random read performance in a bar graph at each queue depth. These 4 KB random graphs are partitioned into high and low queue depths.
The general consensus about mixed workload is 80% reads in client environments and 70% reads for
SATA based devices are half duplex. These devices can read and write separately at the same time and cannot handle both processes at the same time. Products based on the SCSI instruction set (including SAS) are all duplex. These products can read and write at the same time. Full-duplex devices are better in mixed workload environments.
Boot drives perform differently under load conditions; Because the system is constantly reading and writing small pieces of data. When you start a program, the software opens as a stream of read data but also records write data. These processes occur thousands of times per minute.
Second drives used for mass storage change the read-write ratio. They do not log operations, but read and write files "to" and "from" the system as they are transferred. Most secondary drives store data that is sequential have been transferred Movies, music, photo collections, and other media make up a large portion of secondary storage space. In the next section, we look at different transfer ratios. We also look at how sequential data reacts while multiple tasks are occurring in a secondary environment.
Steady state operation is often It is related to organizational workload. Most of the time, this is where I think the situation is stable. Client SSDs are idle most of the time. The TRIM command, garbage collection, and wear-leveling schemes have a chance to clear NAND cells that are held for new persistent writes.
The two images above are about steady state performance. In a client environment, you'll never write 4KB blocks to your SSD for hours. The first diagram shows the second pass and not even the initial pass with clean cells available to absorb the write workload. The second chart is what we are more interested in looking at. This diagram shows what random performance could look like in the worst possible scenario. Ideally, you'll see high IOPS as well as a steady stream of data without too much drift.
There are examples in this regard where steady-state performance data is more relevant. A sequential mixed workload steady state test shows us how a drive behaves after heavy media editing on a secondary device. Since we don't know what each user's routine is like, we show from 100% reading to 0% reading (which is actually 100% writing).
Hard drive testing software is Real
We've used the PCMark 8 benchmark to test the performance of SSD, HDD, and hybrid drives to review Microsoft Office, Adobe Creative Suite, and a selection of popular video games. You can test the system drive or any other known storage space like external drives. Unlike synthetic storage tests, the PCMark 8 benchmark shows real-world performance differences between different storage drives.
When we move past synthetic tests that measure the ultimate performance of each corner, we move on to real software. Our storage results come from Futuremark and are considered part of the PCMark 8 suite. The standard PCMark 8 test is based on a number of software. The way of working is that the software is executed and the traces of input and output (I/O) are recorded. This software then plays back the recorded data on your computer; Just like when you do something in the real world. The aforementioned benchmark also broadcasts data pauses; As the heavy work in the system is stopped. This software is the most advanced test available to check hard drive behavior in the real world.
A standard implementation , gives us a result for each of the tests in the form of service time. Most of the time, these numbers show only small differences between premium and low-end products. This also happens in the real world.
PCMark also gives us a breakdown of average throughput. It shows all the tests. This result shows us a wider range with the total software workloads. Individual findings are misleading; Because they only record a single moment in time, the final output is worth about an hour's worth of work on average.
Apparently, it wasn't enough for Futuremark to publish The best benchmark ever available in the market will suffice. The company strives to further develop the best storage benchmark ever created. The Storage Continuity Test works on a single drive in a multi-step process. For years, we've known that SSDs should be evaluated in 3D. The two-dimensional view only provides a simple picture of the performance status, but the lack of depth in the steady state and the absence of recovery performance are evident in this view.
1. Burn to the drive sequentially up to the maximum reported data capacity with random data. The write size is equal to 256*512=131072
2. Do the writing process one more time (to be sure).
1. Run random size writes between 512x8 and 2048x512 bytes with a random offset for 10 minutes.
2. Run the performance test (one pass only). The result is stored in the secondary findings section with the prefix name degrade_result_X. X is the counter here.
3. Repeat steps one and two for eight times, and in each pass, increase the duration of the random burn to five minutes.
1. Writes of random size between Run 512*8 and 512*2048 bytes at random offsets for the final duration in the parsing phase.
2. Run the functional test (one pass only). The result is stored in the secondary findings section with the prefix steady_result_X. X is a counter here.
3. Repeat the first and second steps five times.
1. The device should be idle for five minutes.
2. Run a functional test (only one pass). Save the result in the secondary findings section with the name recovery_result_X. X is a counter here.
3. Repeat the first and second steps five times.
1. Continuously write to the drive the reported capacity with zero data. The write size is equal to 256*512=131072 bytes. The workload is the same as the standard test. The result is a very long text file with many useful bits of data. We use the overall output of each combined test and overall latency.
Notebook Battery Life
MobileMark 2012 v1.5 is a software-based benchmark that measures usage patterns of business users in a software environment. It reflects like Office and content creation programs. Unlike benchmarks that only measure battery life, MobileMark 2012 measures both battery life and performance at the same time, showing how well a system behaves with a balance between performance and power management.
Meanwhile, we used Bapco's MobileMark 2012 v1.5 to test our notebook's battery life. This company recently released MobileMark 2014 and after solving the few problems of this program, we are finally moving towards a new software. MobileMark 2012 v1.5 software comes with three test scenarios: office productivity, media production and consumption, and Blu-ray. We use the Office Productivity Benchmark exclusively.
We use two separate systems to run MobileMark 2012 v1.5. The first one is the Lenovo T440 notebook that we are testing the SATA 2.5 SSD and HDD. This allows us to benchmark mSATA SSDs.
The second system is a 3rd generation Lenovo X1 Carbon with an M.2 hard drive used to test SATA and PCIe based devices. performance results and notebook battery life are not comparable between the two systems. At this time, we have not found a single notebook that allows us to test all formats on one device.
MobileMark 2012 v1.5 software is installed and/or uses the following 13 programs:
In order to keep the test stable, each notebook needs fresh batteries after 10 tests. On average, a new battery is needed once every two months. To maintain consistent results, we've used Lenovo's six-cell batteries for the T440, as well as Lenovo's internal battery for the X1 Carbon.
When this is done, we end up with two sets of numbers. The first is the per-minute measurement, which tells us how long the notebook has been on. The second is the performance score. In a low power consumption mode, the said notebook reduces the bandwidth and clock rate in several components. SATA bus along with CPU, GPU, DMI link and DRAM degrades less quickly to increase battery life. Our performance rating shows efficiency while available power is a limiting factor.
From time to time we take a picture of the PCB with a camera. We emit heat. We don't do this in every review, and more often than not we include this procedure when a new SSD controller comes on the market. To show the heat generated, we publish two images: one for the drive that was idle for 10 minutes and the other after writing 4KB blocks for 10 minutes.
In some environments, you may not need a solid state drive that can reach 114 degrees Celsius under heavy workloads.
Flash NAND will have the best efficiency in a certain temperature range. However, flash can still accept writes at the higher end of that range, but endurance suffers. Even flash working at high temperatures can cause problems in the long run. NAND consumes energy; So, some heat is generated from this, but most of the thermal energy of an SSD comes from the controller. We looked at the hard drive design to see if the drive developer had enough flash from its processor.
The Last Word
Several exciting changes will affect the hard drive market this year. SSDs are going to get a better performance interface and a simpler command set. Meanwhile, the arrows are also growing. These developments have led to segmentation of the market. While low-end products compete with mechanical storage spaces, high-end products allow new software to shine. The two terms NVMe and 256-bit 3D NAND have been prominent in this field for several years. NVMe is a set of commands that frees NAND from the limitations of Advanced Host Controller Interconnection (AHCI). AHCI was introduced as a registry-level interface for SATA. When SATA was introduced, flash did not exist in the dimensions and densities we see today. At that time, hard drives were still ruling the roost for several decades. Of course, the mechanical nature of these devices limited their performance. NVMe changed the row limit from 32 to 64 thousand; This means that each queue can handle up to 64,000 commands.
NAND flash is also progressing and developing. Optimizations in manufacturing technology were already present in Samsung's first 3D V-NAND. IMFT continued to pursue 3D flash in mid-2015 and it was rumored that we could see 256GB densities in the future. In any case, 1TB SSDs will change to 2TB. However, the cost of manufacturing and development should be such that the customer can afford it.
Also On the flash front, flash expects three bits per cell, also known as TLC. In many charts, the non-branded SSD is shown under the name SMI SM2256, which is a board from Silicon Motion with a new controller that will hit the market in the next few months. This part is designed to support low-cost TLC flash with P/E cycles as low as 500. Advanced LDPC algorithms are expected to extend the life of low-end flash.
New high-end products definitely require tweaks to our testing methods, but the same is true of low-end products. Faster storage is expected to overcome host interface limitations. This means four-cornered PCIe 3.0 Link or 32 Gb/s. The unveiling and release of PCIe 4.0 is not far from the mind. LDPC adapts to changing flash. If an error occurs, the controller goes back to read the flash cells again. This procedure increases latency.
We've seen quite a few cheap TLC flash effects cause performance issues with the Samsung 840 EVO. This 1nm NAND shows signs of repeated read attempts after a few months. If this Samsung device suffers from this problem many times, the mentioned product will not reach the minimum standards.Buy an external hard disk from SafirSoft.
Source: Tomshardware a >