Is it possible to directly plug the M.2 solid-state NAS into a PCIe adapter card? I installed the 128G old disk and ran it for a month. I calculated the three accounts first and then started.

The N100 small host at home is equipped with Feiniu, and one of the M.2 ports on the motherboard has long been occupied by the system disk. Last month I rummaged through a drawer and found a 128G Zhidai PC005. It was an old NVMe disk that I had discarded a few years ago. It was a pity to throw it away, so I thought about attaching a solid-state disk to the NAS as a download disk - the mechanical disk creaked when downloading at night. Is solid state cache worth it? I complained about it in that article. But the motherboard is out of stock. When I searched on Taobao, PCIe adapter cards costing 30 yuan were all over the screen. "NVMe to PCIe x1" and "four disk slots in one pot" were very popular. I placed an order for the most common single-slot adapter card, installed it and ran it for a month. The water inside was deeper than I thought, and the three accounts were laid out first.

first account: The adapter card does not increase speed, but may be a bottleneck.

Many people think that the speed of an NVMe disk is more than 3,000 Mbit/s. On NAS, this number is basically an illusion. My card is a physical port of PCIe x1, and the channel that can be allocated to it on the N100 motherboard is also x1. The theoretical bandwidth of PCIe 3.0 x1 is about 985MB/s. After deducting the protocol overhead, the measured sequential read and write can reach about 800MB/s - sounds good, right? But please note that this is the upper limit of "the disk follows the card". If your disk is rated at 3500MB/s, it will only be a fraction if it is inserted into the card. Only x4 cards with x4 ports can run up to the nominal value of the disk, and a small host like the N100 has a total of 16 PCIe 3.0 lanes. After the network card and SATA controller are divided, there are often only one or two x1 left for expansion.

What's more important is that this bottleneck is mostly invisible on NAS. The Gigabit network port is only 110MB/s when fully loaded, and the 2.5G port is around 280MB/s.—— Do you want 10 Gigabytes? After calculating this in that article, I stopped at 2.5G after going around for a while. In other words, even if the adapter card limits the disk speed to 800MB/s, it is still far from the ceiling of the network port. For roles such as download disks and cache disks, random reading and writing are the focus. It doesn't hurt to cut off the sequential bandwidth. What will really hurt is those who use the disk on the adapter card to frequently move large files. Then you have to make sure that your machine has x4 ports.

There is another piece of trivia: Most cheap cards are "transfers" rather than "bridges". There is no bridge on the card, and the NVMe signal of the disk goes directly to PCIe, so the card itself hardly consumes performance. The bottleneck lies in whether the port you plug it into is x1 or x4. On the other hand, multi-bay cards (the kind with four M.2 cards in one card) must have bridges for PCIe splitting. The compatibility of off-brand bridges is another risk. For home NAS, I recommend directly bypassing them.

second account: Compatible with this level, old motherboards and old firmware are most likely to be stuck.

The problem that the adapter card player will not recognize the disk when it is turned on is the most common problem among adapter card players. I also encountered half of it this time. When it lights up for the first time, Feiniu does not have this disk in the hard disk list at all. lsblk I can't see the nvme device either. Don't panic, put them in order: Check the BIOS to see if the PCIe device is enumerated, but it is not in the system. It is probably because the "pre-boot delay" of the card is not enough - the link training of the old motherboard is slow during cold boot, and the system starts counting devices before the disk is ready. Cheap cards have jumpers or dialing codes to add delay. My card cannot be adjusted, but I figured it out once I restarted it, and it didn't happen again for a month. If it still doesn't work no matter how you restart it, try changing the slot. The channels assigned to the PCIe slot of boards like the N100 are cut out from the chipset. Some slots are reserved for graphics cards, and some ports are only open at half speed. If you insert it in the wrong position, the light will not light up.

The long-term test is whether you can lose the disk after recognizing it. The NVMe disk on the riser card is equivalent to an extra layer of physical contact.: The gold finger, the card slot, and the PCIe gold finger of the card are in contact with each other. Any one of them is oxidized or not plugged in tightly. The symptoms are the same as mine. Solid state disk failure troubleshooting It's the same as what was written in that article - if there is no problem in the list, it will be gone, and the reset error related to nvme in dmesg will refresh the screen. The judgment method in that article is copied here.: Whether the fault is caused by the card or by the slot, the pot can be divided in ten minutes by changing the slot. In addition, the most common offender of APST power saving is on NVMe disks. Small disks such as 128G are inherently weak. If they are inserted into the download slot that has been idle all year round, the probability of power saving is not low. In SSH, nvme get-feature /dev/nvme0 -f 0x0c If you can see the APST status, turn it off if disk drops occur frequently.

I have to mention here a pitfall that many people have fallen into.: Set the NVMe disk as the first boot item and then remove the card. The system is installed on the adapter card disk. If you move the card to another slot or simply pull it out, the machine will not start and the BIOS will still point to the device in that slot. Try to use the motherboard's native M.2 port for the NAS system disk, and use the disk on the adapter card honestly as a data disk or cache disk. If this rule is followed, it can save half of the "cannot boot" tragedy.

The third account: In terms of heat dissipation and power, the adapter card is more picky than the original card.

The location of the M.2 port on the motherboard is generally considered by manufacturers for heat dissipation, so it should be close to the motherboard heat sink or right on the air duct. The riser card stands in the middle of the chassis, and the environment depends on luck. My 128G disk is running naked on the card. There is an 8cm small fan in the chassis that rotates slowly. When downloading, I touch the label surface of the disk with my hand. It is about 55 degrees, which is acceptable. ； But if it is plugged into a corner without air duct and a mechanical disk is pressed on it, it is common for NVMe to trigger frequency reduction when it exceeds 70 degrees, and the speed drop is still slight. The long-term high temperature of the main control is the aging accelerator - how to read the temperature scale? 52 degrees and 58 degrees are two different things I wrote in that article that the threshold of solid state is wider than that of machinery, but don’t run naked just because of its width. It costs more than ten yuan to add an M.2 cooling vest. The cost is not high and the benefits are real. It is especially recommended for old disks with small capacity. Their heat sink configuration is basically zero.

The electricity bill is simple: The single-bay riser card does not receive additional power supply. The 75W of a PCIe slot is more than enough to feed an NVMe disk. Even if you have a full disk, you cannot get 10W. What you need to pay attention to is the half-height and half-length type of card with a SATA power supply interface. The power port is prepared for multi-disk cards, and the single-disk card can be plugged in and left alone. What really counts is "Is this disk worthy of being powered on 24/7?" - When a 128G old disk is used as a download disk, the electricity bill will cost more than ten yuan a year, but its writing life is another matter. How to check how much TBW is left? I wrote it in Should TBW be replaced? In that article, take a look at the old market before taking up the job, and don't let a soon-to-expiry market carry the burden of downloading a large account.

Used it for a month: The value part is indeed valuable

After installation, I pointed the entire download directory of qBittorrent to this disk. After the download was completed, it was automatically moved to the mechanical disk. How I feel after one month: The creaking sound of downloading at night has disappeared (yes, it is the sound of the red disk), random writes of small files such as container logs and download databases have been moved smoothly, and the hibernation of the mechanical disk can finally sleep through the whole sleep - I am here for those habitual criminals who steal disks. The hibernation failure article Called each name one by one. In terms of speed, the download bandwidth is not much. With the 2.5G network port, this disk has no pressure. How to measure the speed so as not to be deceived by the cache layer? Let’s see. Solid state speed test That one will do.

In addition to downloading, this disk also replaces part of the work that was originally done by the cache disk.」: Read and write the pile of broken files in the Docker container. The underlying logic of this arrangement is here Is caching worth it? I wrote the conclusion in that article - the most stable use of solid state on NAS is to serve random small files, not large files that are read occasionally. The NVMe on the adapter card is more than enough for this job, and the 128G capacity is enough for downloading and logging. What can’t be accommodated is your own video hoarding habit, which requires a mechanical disk.

Rollover record and unresolved problems

The overturning point is also explained clearly.: I lost the disk once in the first two weeks. In the morning, I saw the download directory reported as read-only. When I restarted and checked the logs, I found that APST had died from sleep. Solid state disk failure troubleshooting The prescription in that article is exactly right - after turning off the power saver, there has been zero recurrence today. The disk loss is not unexpected news on the adapter card. The extra level of contact increases the probability. Fortunately, the number of troubleshooting methods is exactly the same as the original port, and the log has the final say. In addition, the problem of having to restart once to confirm the disk has not been cured. I occasionally have to wait for it to "think about it" during cold start. This is the ancestral character of cheap cards, so I can't bear it. If you are more serious than me, if you cannot recognize the disk during cold boot, just check the BIOS to see the PCIe enumeration. I wrote the path in the second entry above.

What kind of person is suitable for taking this path? I will collect a list based on my own account.:

End in one sentence: The thirty dollars spent on the adapter card is not about speed, but the opportunity to "turn idle disks into warehouses." For the bandwidth account, first calculate the number of channels, for the compatible account, pay attention to cold start and APST, for the heat dissipation account, spend more than ten yuan and add a vest - all three accounts can be passed, and it is not too late to place an order. I really plan to upgrade the NAS as a whole. The priority of memory and network port is before the adapter card. Those two accounts are Memory upgrade There are ready-made ones in that article.