It's not theoretical, it will just take longer for you to experience the problem so it's something you need not be overly concerned about at the moment.

The point of my commentary in this thread is to try to inform people how operating system memory management actually works so they can make a more informed decision because people may have 8GB of RAM in their computer, read on some website "disabling swap improves performance OMG!" which is patently false, and then have various computer problems after that and not know why.

The fact is that having swap enabled including for you does not in any way decrease performance. In the best case for you swap will not be used or only a very small amount will be used by the OS that has zero performance impact on your system. Disabling swap entirely however will have less of a noticeable impact on someone with an outrageously excessive amount of RAM for what they use the computer for, but an increasingly more likely noticeable impact as their RAM usage to availability ratio increases.

It isn't a case of "disable swap and your world will come to an end tomorrow no matter how much RAM you have", but rather a case of disabling swap provides no benefits at all so there's no reason to disable it. If your system doesn't use swap or only has marginal amount of swap use which is normal after a period of time then there is no consequence whatsoever to having it. If you genuinely have enough RAM for your programs to run then they'll run fine and the OS will never have much need to swap so they'll remain entirely in RAM. The swap file will exist and if an occasion arises where your system has a spike in RAM usage from something, it will be able to handle it. Programs memory leaking can do this, and having swap will buy time to try to kill the offending program once system performance seems to start stuttering rather than reaching an OS triggered OOM condition where the OS kills various apps trying to stop the problem. I've prevented such problems countless times and it has been a life saver on a number of occasions.

But, I'm not trying to convince you to change your own habits in any way either. You do you. I'm just providing fairly watered down info on how memory management works and why disabling swap is almost never a good idea and yields no actual benefits except for a very marginal savings of disk space.

There are people out there who will claim that disabling swap improves performance which is total nonsense, and no OS engineer with experience in memory management would ever agree with that because it is simply not true. If swap is actually causing one's system to become slow for real, then that is an indication that they do in fact need more RAM, but not so they can disable the swap file, but so they have a more proper RAM to swap ratio for the type of software they are running.

What I wonder, is if anyone who disables swap has ever bothered to google for "why disabling swap is bad" or similar terms and try to find a technical article from someone that at least seems to be reputable on the topic and not just some random blogger shill.

I don't think it was ever about performance, more like avoiding unnecessary writes and reads. I do agree with you that the average user with an average PC does need it. Personally, though, I prefer to solve the problem via overabundance of RAM. It's so cheap, that doubling the amount at the first sign of trouble is trivial.

Now, realistically, neither excessive writes, nor running out of memory on a good modern system is a concern. I know I will never run out of RAM and I know that I will replace my SSD a decade before it would start showing signs of wear. At the high end it's just a preference thing.

1) For all that is holy PLEASE do not disable virtual memory. Lord knows I've had so many bizarre game issues be fixed by asking "do you have your page file disabled". Because this causes all sorts of silly issues that are not obvious like a message that says "running out of memory"

2) Your SSD is not going to 'wear down' if you put the paging file on your SSD. Only companies that write literally petabytes of read/writes ever hit the point where the entire drive starts failing. Unless you're running your own Akamai distribution node as a side business, you're never going to hit the kinds of limits to makes this 'unnecessary'

3) If you are REALLY paranoid about your SSD (which again lets be clear isn't a thing that happens) you can always move your paging file to a different drive. The sort of bizarre irony is that if you're system is paging enough where it needs that file, you are better off having it on faster SSD than slower disk.

4) A more reasonable thought process is to reduce disk space usage on your SSD and offset it to a slower drive

Yay, GOG isn't letting me post my full reply, so i have to break it down, peice by piece.

Funny, i've written an OS of my own and I disagree with them. Great, now we have "argument form authority fallacy." Sorry, but a guy who has been decapitated isn't dead due to malaria, just because the guy in a lab coat with a PhD says so.

Except in scenarios where you're actually running out of RAM (the one you're describing). When it has to move things to swap, it will. It prioritizes, so if you have 5 programs open, each eating up a gig of mememory and you've only got 2 gigs of RAM, you're going to see this swap writing and reading all the time, whch I do when i make the mistake of trying to run Skyrim and Firefox (research either the game or what my girlfriend is talking about), Skype (where i'm talking to my girlfriend), and Skyrim (the game i'm playing) at the same time. Sure, in this precise case, i need swap to run all the programs, but it's an extreme case where we see precisely what's going on: the benefits of swap freeing space to cache file operations is not happening, since there's not enough RAM. In the other extreme, it shouldn't happen, since you're not running out of RAM. You're focusing on the middle cases, which aren't as common, anymore, 'cause at the numbers we're now dealing with, the disparity between the edge cases is alot larger. The kicker, though, is that the realistic solution hasn't changed in over 10 years.

But, even if they were: should not your advice then be to always have as much RAM as possible? It's not like there's any reason why the swap file can do things that more RAM can't. Depends on how much RAM is needed. Ah, there's the argument. So, instead, we should disable swap, right? If we disable swap, we run out of RAM, and the OS will actually, then, give us a pretty little notification that says "yo, you need to go buy more RAM, but you can enable swap temporarily to make this message disappear," right? Yes, I agree. I say this should be for the whole system, overall, though. What is it that the game devs say about their RAM usage? RAM is cheap, these days, right? Let's advocate for that approach and see how the customer really feels about these arguments, right? Right now, game devs are trying to say that swap is a magic solution (that way, they keep their "buy more ram" argument to other coders like me who argue for more optimization, but never actually have to present that argument to a customer), hence this post. Minimum specs vs recommended specs vs realistic specs are always a good conversation to have. Right: in this case, we're returning to "let's make the OS do things we're not willing to do ourselves" (which is great in theory, but bad in practice, but was good in practice back in the 80s and 90s). This is why garbage collection is so frequently touted: over commit, data in the program that hasn't been used in hours and, likley, won't be used again will now be in the RAM forever, since no one bothered to free it. Unlike with the stuff above, the OS isn't going to cache this stuff. This all comes back to: we really need to hold programmers accountable. It's not even, really, overcommit that i have a problem with (as you say, it doesn't cause problems) so much as the mentality that it affords: detect ram of system, declare all of it, use it liberally, and pretend the OS will do garbage collection for you. This is precisely the mentality that has java and rust programs performing better than C++: garbage collection. It's not even necessary to do garbage collection, but it makes things easier. Why should we make the OS do this? That only takes more control from the programmer (since the OS devs will eventually be forced to do it), and ultimately eats alot more RAM, since it's not actually happening. It could be that i'm out of RAM, but it's acutally happening in some programs, where the program actually loads the file, frees the file, the loads it again, within the same frame. It's bad optimization, but why bother if the OS is doing all the optimization for you? I can run Skyrim, but not very well. I could also be seeing this happening under the hood because windows doesn't do what you describe. I really don't know which, to be honest. What i do know is, the programmers who work at Bethesda should've seen this ahead of time, as it's not even a fringe case. People have made mods for skyrim to make "optimized high-res textures" to mitigate the problem. However, it's just one case, but a great one exemplifying the mentality that has been plaguing comp. sci. for a very, very long time: "optimization is not my problem." To be fair, i understand why, but there needs to be a point where customers have some sort of backlash against the problems. Until then, we should warn people, so smart people see the problem ahead of time and optimize properly (most likely in open-source where projects are largely volunteer) and the rest are ready for "well, we told you so." Meanwhile, it makes for happy customers of the people that did optimize, which is what i noticed and it convinced me to focus more on optimization. Right, but that instantly became "using the memory." In other words, if i declare a gig of memory, but am only going to use 1MB, i just used 1023MB of RAM and zeroed it (for security reasons) while simultaneously expecting the OS not to actually allocate that until i use it for something else. Remember: with over-commit it doesn't give me the RAM until i write to it, but 0s count as writing. Or have we done something, already, to counteract this that i'm not aware of? I've seen .pch files and such left in "final versions" of products, already. Honestly, i think we should move to a model where debug symbols are stored separately, and if there's an error in the code, your reports need those debug files to comprehend the error reports. Instead of writing "crashed in function X", write a build number and byte number as well as the numbers in the call stack. I know that can be a problem with dynamic allocation, so perhaps some other data can be grabbed, too, leaving them in the main executable is unnecessary, especially in a day and age where single executable/linkable file products have become rare.

Great in theory, not so great in practice, but this is where your swap argument comes into play. I know first hand that reality is quite different, since i'm actally using swap when i play skyrim and still have these problems. If programmers took on optimization accountability, themselves, we could develop reasonable measures to make the task much, much easier, especially with things like games where huge amounts of RAM and disk is being used regularly. Why not hook into file read operations, while playtesting, and have flags for certain reasources that are pulled up every so many frames/seconds so that they're constantly cached? Or what about having a timer? Windows' OS level RAM management is not working for me, even with the swap activated. It's not succeeding at this job, so why are we using it? It's like "why buy a new screwdriver if the company provides us screwdrivers? Sure, the one we have to use is broken, but that's OK."

But, this is not unusual. I remember a similar style arugment many years ago when i argued with devs that everyone should be at least familiar with assembly of their target platform in non-cross-platform projects. Their argument was that compilers are really good at writing optimized assembly, which I had debunked, thoroughly, to which they responded what i was pointing out was "negligible." I counter argued, but that became a matter of "it's not feasible to optimized, as there are deadlines and not enough funds, etc, especially as assembly's too hard," which i responded with showing that assembly is not too difficult (with example code), and then it became an issue of "well that's not readable" or, more often, personal attacks. The most creative counter-argument i got was this dude
who got an program to compress his EXE and said that in the same amount of time it took him to code and grab his EXE compressor, that's how long it took me to code the same program in assembly. When i said that's because he already had an exe compressor available, while i was not aware that such a thing would come into play, hence why i didn't grab one of my own (since i didn't have one handy, since that only saves physical space and not RAM space), then it turned into a personal attack about how his using C++ instead of assembly afforded him the time to have one available.

But, the good news is, compilers, since then, have gotten alot better at optimization (even though it doesn't seem to be used). I'm not sure we can say the same thing in the future about OS level micromanagement of program memory, but we do know all this push towards system APIs (like .net) seems to be for that purpose. My experience makes me skeptical, especially with all this terrible memory management going on in the present.

I remember getting the "RAM is cheap, so customers are always willing to upgrade" argument when arguing with people about taking the time to optimize. Seems like the goal post has moved, then. You're right, though. I don't think a customer should have to. I really do think we should be thinking this stuff through, though, rather than using swapping as a magic bandaide for the problem. I think some sort of program stating what programs are using how much RAM would just as easily solve the problem. "Yo, dude, you can't have 100 tabs open in firefox and expect our game to load the highest res textures at the same time. Our specs assume you are running nothing else at the same time, so adjust them accordingly." I'm sure that wouldn't make the end-user happy, but neither does other solutions. You should be realistic with people, and when enough people make an issue out of it (while being polite to the customer), it's go around the same way as this "don't disable swap" post did. Well, that is a thing, but companies should not be thought police. If i find a game that i've bought that checks for swap, i'll write a bad review, and if possible in that period, demand my money back if swap file was not listed in the minimum requirements. I'm a smart customer, probably with more knowledge of what i'm doing than the people who make the game or whatever. They're effectively telling me that they know better what's best for me than I, myself, do, which is arrogance. This would be included in my negative review of the product. Only pages, not entire programs? And what about when programs revive themselves due to some javascript on the page? Sure, maybe this is why browsers anymore want to use separate instances for each tab, but that is a whole other topic with a whole bunch of other problems (like memory fragmentation). As for the game crashing, well, that's the risk i take for having all these other programs open at the same time. However, most likely time for game to crash for this reason is on "area change," which is also, conveniently, when modern games auto-save. Isn't that interesting? Though, i won't pretend it's actually for RAM reasons. Argument from authority. It's reasonable, however. But one must hold people accountable for whom they choose to read, and they should hold authors accountable for their credentials in cases where it might hold value. However, the intersting thing you're saying is, one should have experience with making a swap memory system before criticizing it. It's almost like "you need to invest in stock in company X before criticizing it." I know, I know, strawman, but it's similar enough. Someone who's invested in something shouldn't be the only people allowed to criticize that something, it causes pathological bias. Of course the guy who got paid to make incandescent light bulbs is going to say LED light bulbs are bad. Of course we expect drug companies to criticize homeopathic medicine, and vice versa. Defrag is automatic on most modern windows installations. IIRC, not necessary for EXT formats, so i'm not sure where linux stands on that. Though, i'll second that static sizing thing: if you're going to play with the devil and use a page file, you really, really don't want it to be fragmented from resizing, and you also will see a performance during "bad times" when you don't have resizing going on, and imagine if you fill up your disk and the page file can't grow.


Where were you when i needed you 10 years ago to make this argument when people said that programmers shouldn't need to optimize?

There are many Windows programs that are coded in a way where they require paging to work stable.

Personally on my main rig I disable paging on my system SSD and all other SSD/HDD except for a 120GB SSD partitioned 32GB for pagefile system managed (I have 16GB RAM).

You have something here, but your response is bad. You're taking agency away from the average person, which is why we see such brain-dead apathy from them on these topics. Why would people ever be invested in learning the truth on matters if it's enough to just take random advice from strangers? This is true if and only if swap isn't used because there's enough RAM. Theory and practice are often two very different things, and I think we've hit the crux of this disagreement. We have different practical experience, and we have vastly different theory, and we're arguing theory, and not practice. In practice, you disable the page file, then re-enable when you don't want to buy RAM. If you let windows do it, it ends up being the microsoft equivalent of apple throttling the CPU "to conserve battery life." In practice, we all know that the average shmuck (who would have trouble following the theoritical arguments) is the kind of shmuck that buys a new computer to upgrade both the RAM and the hard drive. And disk life? But you did actually turn around and throw insults, actually, so it wasn't just enough to propose an opposing viewpoint, but you actually wanted to declare that we didn't know what we were talking about. Hi. I really don't like arguments from authority, but, anyways, hello again. It improves performance, and i do have experience in OS engineering and, by consequence, memory management. Not something i spent alot of time on, but this argument from authority is getting ridiculously old. So what's more noticeable, slowdown or a message box telling you that you need more RAM? Why google for opinions from other experts? Don't get me wrong, i've read arguments from opposition before. They are my opposition, after all. Personally, i don't like the field, you get too many people who think that their professor's word is the omnipotent knowledge of computer gods. They get experience by implementing a few algorithms that they were told, and whammo, qualifications. It's also known that a lot of these people can't even code simple programs that draw triangles out of text for an interview. Employers are constantly complaining about it. Then you have people like me without a degree who don't even bother coding much, anymore, because it's not worth doing a bunch of things that you'll never get paid for since you don't have the sheepskin and 10 years of experience working for a big company. Nah, i'll stick to my random pet projects and do what i think is fun instead of what someone else feels is useful. Maybe the industry will grow up some day, but i gave up waiting for it to.

You really should learn what a page file is before disabling it. If you do know what a page file is, then disable it, you have no excuse not to correlate the actual contents of the message telling you about running out of memory. Not necessarily true, actually. It's not the amount of space, though alot of drives try to distribute the bytes evenly to limit this. it's how many times a particular bit of memory has been written to. It's not just SSDs, but HDDs also have this problem, too. Generally, if we're talking about a mixed SSD and HDD system, i'm more likely prefer more data to be done on SSDs, while more permanent data to be on HDDs. Which one is going to hold the data longer if power outage occurs or something? I know, SSDs don't loose data over the course of a few days, but they do over the course of a few years, while HDDs hold the data longer. For that reason, i try to keep solid state devices powered at least every so often to help ensure the charge has a chance to restore (assuming restoring charge is possible for the type of memory they're using).
Sounds to me like these windows programs are memory hogs, or your environment is hoggish, but i think i'll take the former over the later.

Oh god.. the wall-of-text are increasing exponentially O_o

Honestly, it's to the point where i'm thinking of temporarily moving this over to PHPBB on my server (which i'd have to set up) or encrypting it into a lossless png or something. You can't even get our arguments into 1 post anymore.

EDIT: Just for fun, i just found a fun way to do it in linux, and now you're looking at some source code i did.

Hey, stupid question, what type of memory is better for ssd? if I constantly overwrite files on it? I want to transfer the system to ssd. is there a big difference between TLC, V-Nand and 3D TLC? (example) Thanks in advance.

This is an old thread and completly false imho.
The main difference is price and speed and how new the technology is. Imho just buy which is okay for your wallet the speed isn't that different between them. Jumping from outdated hdd to sdd is huge no matter which one you chose.
And as for reliability if the constant write is only the system + some apps then all these are good enough.
For server or vid editing i dunno maybe there are more expensive ssd for those.

Type of memory is totally irrelevant as SSD's doesn't communicate directly with the memory. Harddrive is considered long term memory. DDR RAM is considered short term memory. Transferring data from an SSD to RAM involves several parts, including several controllers/firmwares (also adding different drivers and the OS itself). Asking what type of memory you should get is only dependent on primarily the mamaboard, and then the CPU secondary.

As for SSD and the page file: It simply isn't relevant anymore as it once was.

1. 1. gen SSD's were seriously more prone to errors, couldn't data very long and had low storage capacity. Even though SSDs doesn't last as long as magnetic drives, you should change harddrive once in a while anyway long before they wear out. I've changed harddrive for others that waited to the last minute and then the amount of corruption made the disk unreadable.

Buy a cheap USB-2-SATA cable and use that "old" SSD as a "usb-pen" that is guaranteed to be at least 5x-10x faster than a normal pendrive.

2. Paging or swapping data between RAM and harddrive have getting better (and smaller).

3. The amount of writing when having more RAM than you need is so little it just doesn't matter in the grand scheme. There's more to gain if you put temp and other things into a virtual drive in memory if that's a concern, especially if you compress and decompress a lot (like when modding), or testing/installing/uninstalling a lot.

4. There's no real tangible effect having it disabled other than gaining some space. That's like arguing that you can see 1 fps less or more in games.

5. If you want to disable it - then go for it. Just remember that if/when you get some instability issues later, and you've just added one more step in the process-of-elimination-list. However, there rarely are any issues compared to in the Win9x/XP days.

6. And yes - expect every software you use is non-optimized, and like skeletonbow indirectly pointed out, disabling it on a server that is on 24/7 is a bad idea (at least on Windows).

I personally have 16GB DDR4 but I still have the page file on (fixed size) on a secondary drive just in case, and it doesn't give me any noticeable performance gain by turning it off, so I don't bother with it. More RAM, CPU and mamaboard speed, and having the software/game on an SSD is massively more relevant.

TL;DR - I'm arguing that it isn't relevant anymore (and if you have it on at least set a fixed size).

PS: Never diminish yourself by declaring you're stupid, is a noob, or has a stupid question. None knows all there is, and there's progress in asking ;)

Is there a post you're not willing to give bad advice on?

Well, I'm afraid he did not even read nor understand my post.