eMMC vs nvme SSD vs SATA hat vs SATA usb

#1

This is my second eMMC that has been gone in a short time. The first one became unusable in ~3 weeks and this one (the new one) wore off in a month. Although i can still read it, or i think i can read the contents, it is still a disaster if you need to redo everything repeatedly. I am sure it was not completely full, i had 4GB left, but suddenly the controller assumed it was filled up. “It was at this moment i knew i was f* up”. :laughing:

Filesystem      Size  Used Avail Use% Mounted on
/dev/mmcblk1p2   28G   28G     0 100% /
devtmpfs        4.0M     0  4.0M   0% /dev
tmpfs           7.7G     0  7.7G   0% /dev/shm
tmpfs           3.1G   19M  3.1G   1% /run
tmpfs           5.0M  4.0K  5.0M   1% /run/lock
tmpfs           1.6G   44K  1.6G   1% /run/user/109
tmpfs           1.6G   44K  1.6G   1% /run/user/1000
/dev/mmcblk1p1  511M  122M  390M  24% /boot

This Foresee chip (or the controller) can’t really be trusted. Taking apart the usual “do a backup”, and “cross compile in your host computer”, “buy cheap, buy twice”, etc, etc, i would like to know how good is your experience with the current BSP kernel with nvm2 SSD, SATA hat, or SATA USB.

Please, rate the reliability, speed, lifetime expectation, and model/maker/size you currently use.
I know it is a relative question but consider it for heavy usage.

#2

Just out of curiosity. You do know eMMC is used especially only for reading right? It is not meant for constant writing to it like how a SSD is able to.

And looking at your df -h command i see your whole OS being put on it…

Isn’t a NVME SSD a better option? (I use m.2 Samsung Evo plus 970 2TB)

Been using Samsung SSD in almost any device of mine (mSATA 860 1TB (router Turris Omnia), mSATA 850 500GB(desktop), SATA 850 250GB, 500GB(Proxmox server), m.2 NGFF 860 Evo 2TB(raspberry pi 4) and the newest being 970 evo plus 2TB in de Rock 5B)

#3

Yes, i do know. But it was about choice. I had two options, SD card or eMMC. SD card is slow and gets slower when is ~80% full. I have the OS in eMMC but it does not do journaling on the disk, so few files are written to the disk, but i do builds.

That’s not the point. Having a 2TB NVME SSD when you only have 50GB of data in it does give you the feeling you are safe.

In your experience, do you have any SSD with like 10% free space left and still usable if you delete files to free 50% ?

#4

Well in my case i tend to use it the Rock 5B as a low power server to download also newgroups/torrents, PXE server for all kind of images. I dont want to constantly watch if there is no free space left anymore. I am planning to automate things more in the near future, so data gets deleted and is no longer needed. If there is something that i hate is lack of space.

I try not to reach that limit to be honest. That is why i try to have as much space as possible. I am a data hoarder. First step towards change is admitting you are a data hoarder :smile:

What i usually try to do if i have a lot of space left on certain disk, is to use the remaining space to use it as part of a fileserver kind of thing. For example, my Proxmox server OS is installed on a 250 GB 850 Evo and i want to use maybe like 200 GB as a receiver of data from my other devices on my network, because i plan to upgrade my network devices in the near future to 10 Gbit. As SSD --> 10 Gbit network --> SSD is much faster, than SSD --> 10 Gbit network --> spinning sata drive (max 100-200 MB writespeeds). When data is over on that SATA SSD, using script to move all the data to my RAIDZ2 (spinning drives) and automatically shutdown after its done.

At work we do have very often that customers virtualdisk is only 8%-2% space is left. Off course we use LVM to expand the drives. In some cases we just do a simple apt remove to remove all the old kernels and the 8% left goes directly to maybe like 50-60% or so (when the total size of the vda is 20 GB lets say). In cases that the server is property of the customer it self, then in such cases new harddrives are being ordered and put in the LVM setup of the customer in the datacenter itself.

#5

That’s a real-life example. Is this already happening and for how long have you been doing this with that SSD?

#6

No its not yet happening. 2 reasons.

  • I haven’t upgraded my network devices yet (network device in the Proxmox is 1 Gbit), cables i have already upgraded to cat 8 (to be future proof :P), but my switch is still also a 1 Gbit switch.

  • (laziness yes i said it…) I haven’t created the script (which off course is easy). cp -r /directory/* /raidz2/directory_z/ && shutdown. Using it in a tmux session on my proxmox server, i dont have to keep my terminal conneciton open and can just go and sleep :), while its doing its job.

  • [Bonus reason] I currently use the Raspberry Pi 4 (8 GB) to transfer files, however it has sort of a bug which only gives a transfer speed of max 60 MB/s over the network. It’s not the problem of scp if you might think, because on my other devices i do make speeds of up to even 120 MB/s.

The raspberry pi 4 uses M.2 NGFF (SATA) [Samsung 860 EVO m.2 2TB] through USB 3.0 (Argon One m.2 case). So it should be able to give very high transfer speeds.

Btw even though its a TLC, it should have a VERY VERY LONG lifespan. I also try to not use about 20 GB of space to leave it blank so if more cells die, that “spare” space that i am not using can be used for it.

#7

Waiting to hear from you how long your 250 GB 850 Evo will last. :slightly_smiling_face:

#8

I was just searching a bit more. The 850 evo SATA (250 GB) has Total Bytes Written (TBW) (75 TB) with 41,1GB each day.

I will maybe lets say write 200-600 GB or lets say even 1 TB in a year…well 365 days x 41,1GB = 15.001 GB. So it will take a LONG time before it dies because of the amount of writes.

The Samsung Evo plus 970 has a total of 1,2PB with 657,5GB each day. How am i able to make such amount of writes each day for a low power server?

#9

Well, if you read here and there on internet, people estimate eMMC lifespan about 4.75 years. :laughing:

#10

Yes as it depends on what you are doing with your eMMC. I mean you are almost the pimp of that poor eMMC having soo many customers each day will make her indeed retire very soon. Go with a more robust wh*re instead of such a weak one.

Could it be that the eMMC is just of bad quality? While Samsung SSDs as far as i have experienced are really very good SSDs. My to go SSDs for more than 10 years. Never has even 1 drive stopped working in all those years.

#11

Indeed. What i was doing is like:

  • clone a repo
  • edit some file
  • build it, test it.
  • edit, rebuild, test. Edit, rebuild, test.
  • some times i do a clean and a complete rebuild.

Sounds like an everyday programmers life.
But good to hear Samsung SSD quality. Thanks for reassuring me of that.

#12

I’m not sure what a rebuild exactly does in your case, but that sure is writing it to the eMMC. Those tasks are not meant to be done on a eMMC. Just think about a router. If the firmware gets updated (it writes to the eMMC (maybe a max of once in 1-2 years). The syslogs of all the connections of for example wifi devices that have been connected are never written to the eMMC, but rather to the memory. After a reboot everything is lost.

#13

Maybe the eMMC connector plays a role here, a bad role.
I will move my stuff to SSD as soon i can fix my kernel to switch to 9V or just buy a dumb PD.
That’s going to take some time.