Are you a wizard or something? Since when does a Pi have PCIe slots?
1 Like
Thats what im asking, it seems its not
Unless you want to pay a massive performance penalty. If someone has done this pls tell me how.
The pine64 has pcie OP.
And what is your âgeneral feelingâ based on?
1 Like
You can limit the amount of compressed data via sysfs. To let zRAM go sporadic is a very bad idea. On a device with 3gb of RAM, youâre only going to want to use 512mb. And thatâs a standard in general.
The purpose of zRAM is to create a compressed instance in RAM. It still relies on whatever swappiness is set to in order to determine how often it is used. Itâs more beneficial (even in modern systems) as itâs much easier for RAM to decompress an instance than it is for the device to decompress an instance on a storage device. Thereâs no doubt cons to it: your zRAM instance is limited by how much RAM you have, depending on how much you set aside for zRAM, it can carve out a lot of usable RAM, and, at least for the meantime, tasks donât have much priority (rather or not to compress a task is pretty random in general, swap shares this issue as well).
Swap comes with itâs own issues: youâre limited to whatever interface is being used by the partition, the defaults are highly ineffective on modern systems, and the device being swapped to suffers from a diminished life span. Not to mention, embedded devices can actually suffer in terms of battery life as well as it is much more conservative to compress via RAM.
Also, Iâd argue that zRAMs only purpose isnât just on older devices. Google deploys it on current Chromebooks, itâs used by phone manufacturers (a la El Goog) that know their shit, and as someone that compiles the Linux kernel frequently, I find it much more beneficial. Devices with large amounts of RAM will also see a much larger return here too (granted, devices with heavy amounts of RAM often have amazing RAID arrays where the difference between the two become less apparent).
Your analysis of the history of zRAM was actually pretty interesting as I was unaware of its initial usecase.
Anyway, hope youâre having an amazing day. If you have anything to add, then please let me know.
Believe it or not, using a swap space too big is actually harmful to the deviceâs performance. The amount used should solely be based off the usecase of the device. If the medium is too big and is written to often (especially with a default swappiness of 60) then you risk a lot of performance degradation.
And thatâs another thing. I donât think itâs quite understood what the value of swappiness actually does and how to properly configure it. Itâs often assumed that with a lower value, the disk gets written to less and vice versa. But if you have too low of a value then it defeats the purpose entirely. Swappiness also controls how often PAGES ARE DROPPED FROM CACHE. This can be severely bad if a lower value as if RAM is saturated than only foreground tasks are accounted for. This really hurts performance and this is why, on occasion, your computer can seem so sluggish when working on something intensive. Itâs a really important concept to keep in mind when messing with the value and thatâs why the default is so high (60).
Hope I helped in some way lol
There being no practical benefit to swap partitions, and swapfiles are more convenient.
The biggest drawback to a swapfile over a partition (at least in my case) is the lack of BTRFS support. Thereâs also bugs here and there that arise from the use of swapfiles.
Edit: you can use a loop device with BTRFS as per Archâs wiki but it heavily degrades performance
1 Like
That depends. In some implementations swap-files can get fragmented which degrades performance. Also, if you are using full-disk-encryption, hibernation to a swap-file is much more cumbersome.
So in reality, the downsides over-weigh the benefits in my opinion.
1 Like
Doesnât work with BTRFS, true enough. No bugs of note with EXT3.
Fragmentation isnât a real concern as you shouldnât be actively swapping anyway.
Hibernating with LUKS was fixed in 2018.
That sounds like âABS not working in your car is not an issue as you shouldnât be actively needing it anywaysâ. Sure, you shouldnât be swapping much but if you do, it should work, at least as well as a swap partition in this case.
I didnât mean that it does not work but itâs more of a hassle to set up and again I donât really see a reason why you need it to be a file if you cannot use empty swap space anyways.
You canât avoid needing brakes, you can avoid swapping.
Then why do you even have swap-space, if you arenât planning on using it? 
To avoid OOMkiller and a crash if everything goes to hell.
Because itâs clicked to default on the install⌠duh.
1 Like
swap partitions and swap space files serve the purpose of temporary storage in the event of memory overrun exceeding the amount of system memory.
swap partition setup is built in to most distros by default. and often is customize-able with partition management software.
however with many machines made today they have abundant amounts of install-able memory most often the memory can easily exceed the required amount even under heavy loading, so in many cases swap partition or space is not needed.
but all this depends on the file system requirements and graphics rendering as well.
research the requirements of your OS and whats needed for the programs make up a list of pros and cons, and decide what you want to use.
I have bin running without a swap partition for years, but i finally found a situation where i wish i had a swap:
I was firing up a wm for windows. i wanted to give it 8gb of ram, If i had a swap it would have suspended somethings to the swap giving me instant memory access for the wm, Instead of waiting for the system to free up the memory.
2 Likes
The RPi 2/3 is a 1Gb device with shared memory for the GPU, ie free RAM is therefore < 1Gb.
What I did not mention in OP is:
a) I develop GCC projects natively
b) the RPi is also my daily use âdesktopâ, so I often use web browser
c) I am limited to a 12v solar power system
because of c) I cant run large a) projects like build Kernel, GCC and BinUtils. The OP is an atempt to fix that problem on 1Gb devices. Last week the RPi 4 was released which has upto 4G ram, USB3 over PCIE - the problem is mostly solved, the device is 5x faster that the RPi 3B+.
On 1Gb system, the use of an Optane drive is not for the raw speed of the device, but the longevity of the devices write degradation. The 218Gb product has 5x the write life of the 56Gb device, which should allow for maybe 5 years + of everyday use in both a) and b) cases with a swap partition.
On RPi 2/3, Yes the speed of transfers is still limited to the speed of the USB2 over the shared USB2 bus. But the improved swap partition write life means NOT having to upgrade the boot sd-card every 12 months (due to write weardown), AND not suffering the sd-card write slowdowns (which increase with Kernel swaps).
Yes this can still be combined with ZRAM, but a 218Gb Optane can safely support an 8Gb swap partition that will easily last 5+ years if the rest of the drive is only sparingly use (ie not regularly), which directly affects use case a) AND c).
Anyway, as mentioned earlier, the release of a 4G RPi 4 has negated most of these issues, while still allowing 32bit Raspbian usage. A regular high speed SSD (like the recent Adata review by Wendall) will now serve the same purpose, without the extra cost, but most of the benefits.
( and yes, the use of a USB2 compatible PCIE M.2 housing was used )
2 Likes
In conclusion, is there any reason to be using Optane on Linux if you can afford more RAM and youâre running AMD? If not, then which SSD is top? Iâm not exactly clear on the Phoronix benchmark consensus.