1. Current Location: Home >  NAS >  The SSD is rated at 560MB/s. Whose problem is it that cannot copy large files?

The SSD is rated at 560MB/s. Whose problem is it that cannot copy large files?

Comparison of nominal speed of SSD and external buffer speed: Measured picture of slowdown after SLC cache is full

The cause of the incident was that a colleague bought a new SC001 512G. The box read 560MB/s and wrote 520MB/s. He plugged in the mobile hard disk box and poured movies into the NAS. He turned on the 2.5G network card and fully expected to finish a Blu-ray collection in ten minutes. However, he stared at the progress bar all night. It was more than 300 in the first few minutes. Then it went slower and slower, and finally it stopped at around 120. He came the next day and asked me if I had bought a fake disk. I took his disk and tried it on my own Feiniu NAS - there was nothing wrong with the disk, but he took the number on the box for granted. This article will explain clearly the accounts of our actual measurement.

Let’s distinguish clearly first: The number on the box is from the disk, not your copy.

The nominal 560MB/s is the ceiling that the disk can reach under optimal conditions.: Sequential read, full disk idle, single-threaded large block. When you actually copy files, you have to go through several doors with upper limits. The first is the interface: The 6Gbps bandwidth of SATA3.0 is theoretically 750MB/s. After excluding protocol overhead, it can actually reach a maximum of 530 to 560MB/s. Therefore, no matter how beautiful the nominal number of SATA disks is, the physical upper limit is there. The second channel is the network port: Gigabit network port theoretically 125MB/s, actual 110 to 115MB/s ; The theoretical 2.5G port is 312MB/s, and the actual is around 280MB/s. My colleague's NAS is 2.5G, so this door is not a bottleneck. The third way is to interpose oneself - and the gate of intervening oneself has a mechanism.

The mechanism is the SLC cache. Nowadays, mainstream TLC disks will simulate part of the capacity into SLC mode in order to achieve good results.: Data is first poured into this high-speed area, with a speed of four to five hundred MB/s. ; When the high-speed area is full, the master controller slowly moves it to the real TLC area. This action is called folding. When folding occurs and you continue to write, the speed drops to the true level of the disk, commonly known as slow external speed. For a 512G disk, the SLC cache is generally 50 or 60 GB. My colleague's Blu-ray collection is more than 200 gigabytes. The first 50 gigabytes run smoothly, and then the whole process slows down to about 120 gigabytes. That's how it matches. I've hit the same wall on the download disk before. The speed test article The complete process of falling from 230 to 80 is the same truth.

I connected this 512G to Feiniu and let you see the actual test.

Explain the test environment clearly: Feiniu NAS, N100 small host, 2.5G network port is directly connected to the 2.5G port of my desktop. I ran iperf3 first to confirm that the intranet bandwidth is 280MB/s and there is no problem. How to run it? The article on intranet speed test Wrote in. The disk was attached to the NAS using a USB hard disk enclosure, formatted to ext4, and then a 243G mixed-editing material package was copied from the desktop computer, and the speed was recorded throughout the process.

The segmentation of the results is obvious: The speed of the first 48G is around 280MB/s - this wall is held up by the 2.5G network port and USB bridge chip, and the SLC cache of the disk is not full yet. ; After the 48G, it starts to fall down in steps, 200, 170, 140 ; Starting from around 70G, it stabilized around 125MB/s and stayed there until the end. This 125MB/s is the slow external writing speed of the 512G version of this disk, which is 75% lower than the nominal 520MB/s. In the middle, I paused once and then resumed the transfer, but the speed did not return. This means that the cache cannot be freed up after a short break - the main control folding speed cannot keep up with your writing speed, and slowing down is the norm.

I conveniently made up a set of read comparisons.: Copying the same package from the NAS to the desktop ran at 280MB/s without any loss. Because the SLC cache does not add much to reading, sequential reading is the original strength of TLC, and SATA disks can run close to the upper limit of the interface. So "this game is slow" is only half right.: It's not slow to read, it's just slow to write externally.

The slow speed is more or less normal. I turned over a few discs.

How much the buffer drops depends on two things.: Particle type and whether there is independent cache. TLC disks with DRAM cache can generally achieve one or two hundred MB/s outside of buffering. ; TLC has no DRAM (which is often said to have no buffer, QLC is even worse), and the buffer can drop to 70, 80 or even 40 or 50. How many plates do I have on hand to check the account?: That piece of hardware SC001 512G buffer 125MB/s ; The old 480G disk in my NAS has a buffer speed of 82MB/s (the download bit has been running for a year and a half, and the account is at The speed test article ) ; My cousin's 1TB QLC has an actual measured speed of less than 60MB/s. He also wondered why it was not much faster than a mechanical disk. What they have in common is: The speed figures on the product page are all results in the SLC cache area.

How to judge whether a disk is slow? The product page will not say "I have no cache", it depends on the main control model or the teardown post. There is a rough judgment method: The cheapest one with the same capacity, and the one that emphasizes "HMB cache-free design" on the page, most likely has no cache disk. HMB borrows a bit from the memory and uses it as a cache. It works on desktop computers. However, in bridged environments such as NAS and USB hard disk enclosures, HMB often does not take effect, which is equivalent to running naked. My attitude towards allocating disks to NAS is very clear.: Don’t save dozens of dollars on cache bits. Buy one with DRAM. If the cache is gone, the cache will be gone as well. That’s really useless. The article about cache disk disappearing The pit is still warm.

The correct way to copy large files, four items in order

After knowing the principle, copying large files can be planned. My order is this: First, ask yourself whether this file really needs to go through the network - if the desktop and NAS are physically next to each other, copy the USB direct disk and then plug it back in. SATA direct connection 560MB/s is twice as fast as the 2.5G network port and five times faster than the Gigabit network port. The four-way actual test article There are even slower wireless accounts. Second, the network must be used and there is only a Gigabit port. The slow external speed of the disk is not used at all. The 110MB/s ceiling covers the gap between all disks. At this time, the money should be spent on the network port instead of the disk. I will upgrade to 2.5G. Gigabit 2.5G article Calculate it. Third, there is a 2.5G eloquent speaker. At this time, the external speed is your real speed. Check the parameters before buying. Fourth, if there are dozens of gigabytes that need to be copied frequently, SMB multi-channel can raise the upper limit. The actual test of dual network port binding is The multi-channel article Here, the NAS is currently copying large files at 260MB/s.

There is another pitfall that is easily overlooked: temperature. When buffering external writes, the main control continues to be under high load. When the NVMe disk temperature rises, it will automatically slow down. How to read the red lines of 52 degrees and 58 degrees? The temperature article Wrote in. SATA disks generate little heat, so generally you don’t need to worry about them, but if they are packed densely in a hard disk cabinet, you should leave some gaps for them in summer.

Ask yourself three questions before buying

This time I was helping my colleagues read the market, and I summarized "how to read the nominal speed" into three questions. Once you answer the question, you won’t make the wrong purchase.:

question If you can't answer What does the answer affect?
How big is this SLC cache? Read the review and teardown, the official page doesn’t write about it Determine how many G can run at the nominal speed
How much is written outside the cache Same as above, focusing on the actual measurement of 1TB capacity Determines the true speed of copying large files
Is there independent DRAM? Look at the main control model or the words "HMB" Determines random performance and cache bit stability

My colleague didn’t return the product in the end. His disk was later installed into a NAS as an application disk, and Docker and virtual machines were placed on it. It read more and wrote less, and the external buffer speed was untouchable at all. It was very comfortable to use. He learned from me when it came to copying large files.: Take an old 512G and put it into the hard disk box as a "transport tray", plug it in directly via USB, and get 560MB/s in one pass. I didn’t buy the wrong disk, but each disk has its own way of playing - the words "nominal speed" are just for fun, but if you take it seriously, you will lose.

Read More


Copyright Notice Scan to read on mobile
All Rights Reserved: 《SHUNOT》 => 《The SSD is rated at 560MB/s. Whose problem is it that cannot copy large files?》
Article URL: https://www.shunot.com/en/nas/1579.html
Unless otherwise stated, all articles are original by 《SHUNOT》. Reposting is welcome! Please indicate the original URL when reposting, thank you.

Contact Us

Online Consultation: Click here to send me a message

WeChat ID: master_135

Scan to follow