Why not just keep swap on system partition, as is the default?

AFAIK, moving the swap file is only useful if your system has one disk that is faster than the other, and your OS is not on the fastest disk.

yawetaG wrote:

AFAIK, moving the swap file is only useful if your system has one disk that is faster than the other, and your OS is not on the fastest disk.

Wouldn't it be better to have the swap file on a different disk even if the drives were the same speed? That way, Windows would be able to read/write the boot drive at the same time it is reading/writing the swap drive?

As for being on a different partition on the same disk.. that is just going to end up making it slower as the heads are most likely going to have to move back and forth more to access the swap file.

cyclone3d wrote:

Wouldn't it be better to have the swap file on a different disk even if the drives were the same speed? That way, Windows would be able to read/write the boot drive at the same time it is reading/writing the swap drive?

i had such a setup once - no difference save for the most intensive action that did not fit in memory

cyclone3d wrote:

As for being on a different partition on the same disk.. that is just going to end up making it slower as the heads are most likely going to have to move back and forth more to access the swap file.

also had such a setup once with win95 - it was extremely starved for resourced - cpu-memory-swap speed;
a separte 128mb swap partition placed right behind a very small partition just for windows almost made my stuff playable - the frames were displayed in the same frequency as disk activity occured

has anyone ever checked what swap space wants more - linear speed or access time?

mrau wrote:

has anyone ever checked what swap space wants more - linear speed or access time?

Not on Win 9x, but on Win 3.x I noticed that Windows liked to create a fixed, non-movable swap area that was contingous (386spart.par; a series of adjacent sectors rather than a "file").
Programs like Defrag didn't touch it by default. In 386 Enhanced menu, it also was possible to select "temporary" swap file which had worse performance than the "permanent" one.
That being said, the performance difference could also be related to the fact that "temporary" used MS-DOS for access, while "permanent" was either accessed via BIOS or FastDisk (Win3.1's own HDD driver).

Win 98SE goes a bit further and can execute programs straight from swapfile.
.. if they were properly aligned to 4KiB boundaries.

That's one of the major contrasts to the earlier Win 95, which was rather primitive in this respect.

Edit: I forgot - on Win 3.1x, there also another type of swap file, win386.swp, often located in the Windows directory.
That's the temporary swap file, I believe, which gets into action whenever Windows needs to fall back to MS-DOS for HDD access.

cyclone3d wrote:

yawetaG wrote:

AFAIK, moving the swap file is only useful if your system has one disk that is faster than the other, and your OS is not on the fastest disk.

Wouldn't it be better to have the swap file on a different disk even if the drives were the same speed? That way, Windows would be able to read/write the boot drive at the same time it is reading/writing the swap drive?

As for being on a different partition on the same disk.. that is just going to end up making it slower as the heads are most likely going to have to move back and forth more to access the swap file.

Completely correct.
When the system is thrashing for a while, it makes BIG difference when one has swap on a separate drive.
With swap on the system drive, it usually results in very time-consuming seeks over long distances from the beginning to the end of the HDD, and vice versa.
With swap on same drive, you'll have many seeks near the maximum seek time, with it on separate drive, the seeks will be nearer to the minimum seek time.
With the drives I use for my retro computers the maximum seek time is 8ms, the minimum seek (track-to-track) 0.2ms.
So with swap on separate drive, the thrashing will be finished up to 40 times faster.

Out of curiosity, can you put the swap file on a CF card in a IDE CF card reader, or will that wear out the CF card really quickly?

yawetaG wrote:

Out of curiosity, can you put the swap file on a CF card in a IDE CF card reader, or will that wear out the CF card really quickly?

You could, how fast it wears will depend on how often you really use the PC, Keep in mind this isn't your daily driver anymore so could spend weeks even months unused.

Another option along the same lines is install more Ram and create a ram disk at put the swap file there.
Some people might say that's counter productive as the swap file is somewhat a ram overflow, but really these machines aren't running 1/2 dozen programs at the same time anymore so workload is much lighter then it was back in the day. RAM is cheap. so the swap file is more for just keeping windows happy nowadays

chinny22 wrote:

RAM is cheap. so the swap file is more for just keeping windows happy nowadays

I think the same. Some programs like Photoshop Elements require virtual memory, though.
In an ideal (retro-) world, we would have got RAM-based HDDs by now for our vintage rigs (real RAM disks, so to say).
They would interface like IDE or CF cards would do, but would hold 256 to 1024MiB of RAM.

Jo22 wrote:

In an ideal (retro-) world, we would have got RAM-based HDDs by now for our vintage rigs (real RAM disks, so to say).
They would interface like IDE or CF cards would do, but would hold 256 to 1024MiB of RAM.

at least one such device exists; but i cant remember the name; also size is smaller than what you mention

Swap file in an out of the way partition on the same disk is the worst possible option, as it ensures seeks will be long - better to use an out of the way partition for rarely touched stuff

mrau wrote:

Jo22 wrote:

In an ideal (retro-) world, we would have got RAM-based HDDs by now for our vintage rigs (real RAM disks, so to say).
They would interface like IDE or CF cards would do, but would hold 256 to 1024MiB of RAM.

at least one such device exists; but i cant remember the name; also size is smaller than what you mention

Are you thinking of the Gigabyte i-RAM?
https://en.m.wikipedia.org/wiki/I-RAM
Maximum capacity was 4GB. It only came out in SATA format and was discontinued fairly quickly.
I wonder how it'd bench against an Optane SSD for latency, obviously it'd be owned for bandwidth.
There have been many enterprise RAM disks but I don't know of any native IDE options. I'd guess that just having 512MB RAM would work out cheaper even if there was.

exactly - i remembered it as tens or hundreds of megs - must eat less butter

i suppose todays ssds in m.2 format are just unbeatable - but the interface is a problem; even if that is managed (i believe there were pci cards for this) im not sure how much power this requires from the host to address m.2

mrau wrote:

exactly - i remembered it as tens or hundreds of megs - must eat less butter

i suppose todays ssds in m.2 format are just unbeatable - but the interface is a problem; even if that is managed (i believe there were pci cards for this) im not sure how much power this requires from the host to address m.2

I think most people who used the i-RAM didn't max out the storage due to the cost, that might explain it.

I was thinking of the full PCI-E Optane 900p as the beast of access time but maybe the m.2 flavors could still work out better than the i-RAM. There are m.2 to IDE adapters but I don't know how efficient the controllers are.

Yes, I remember the I-RAM. However, there also was a RAM-based floppy drive once, too. Must have been in the early to mid 90s.
It had 1MiB of capacity and interfaced just like a normal floppy drive.

Considering all the cheap RAM chips in the range between 64MiB to 1024MiB, there must be a possibility to create a real RAM-Based HDD cheaply.
It doesn't have to be hi-tec, after all. You find lots of old PS/2 and SD-RAM modules in old computers at the road side.

With something like an FPGA or simila device, it must be possible to turn these RAM modules into something useful.
Latency is the most interesting part- Even an old RAM chip operates in a 10 to 120ns time frame, whereas SD cards take about 1 to 0,1ms until they respond.

There were many I-RAM type products in the server market. Some were on PCI cards, others had SCSI or FC interfaces - You just don't see them out there in the wild though. I guess the kind of places that had to price no object budgets for them don't exactly list their old gear on eBay.

That's likely true, but I believe *we* don't need that much memory.
We can go much lower and still be satisfied with low-tec solutions-

A classic 386/486 PC will be fine with a 512MiB "RAM disk" (CHS limit, btw) that interfaces via IDE or ISA (via glue logic).
And unlike, the i-RAM, we don't require super-fast DDR3/DDR4 RAM either.

A simple old SO-DIMM module from an old laptop found on the roadside will do just as fine.
Or any other SD-RAM / DDR cheap module (for example).
Heck, even SIMMs could be used for this. PS/2 SIMMs even give reasonable capacity. 😁

What makes a true "RAM disk" so superb is the tiny access time (instant access) and the lack of fragmention issues (no jerkyness due to internal houskeeping).
Unlike Flash media (SSD, CF, SD), a RAM chip doesn't really wear out and a true, complete reset of all cells is also possible.