[masklinn]

> It would be nice if the intro had a brief explanation of why a disk needs to be divided into blocks.

One reason is that HDDs simply don't have a byte-wise resolution, so there's little point talking to HDDs in sub-sector units. Sectors are usually 512 bytes to 4k.

A second reason is being able to simply address the drive. Using 32b indices, if you index bytewise you're limited to 4GB which was available in the early 90s. With 512 bytes blocks, you get an addressing capacity of 2TB, and with 4k blocks (the AF format), you get 16TB. In fact I remember 'round the late 90s / early aught we'd reformat our drives using higher-size blocks because the base couldn't see the entire thing.

[jagged-chisel]

> HDDs simply don't have a byte-wise resolution

Sufficient explanation for code. Now why is it that disks lack byte-wise resolution?

[cesarb]

> Now why is it that disks lack byte-wise resolution?

Overhead.

Each disk sector is not 512 bytes (or 4096 bytes in modern disks), it's a bit more. It needs at least the sector number (to detect when the head is flying over the correct sector), a CRC or error-correcting code, a gap of several hundred bits before the next sector (so writing to one sector won't overwrite the next one), and a preamble with an alternating pattern (to synchronize the analog to digital conversion). All this overhead would be a huge waste if the addressable unit were a single byte.

[avianlyric]

Cost, more resolution means more transistors, address lines, and protocol overhead (your memory address take up space too).

You could build a HDD with byte-wise resolution, which did a whole load of clever things internally to handle error correction (modern HDDs have far more complex error correction that just block wise checksums, they’re a piece of technological magic and analogue electronics).

But sure a device would cost far more, and offer no real benefits. Especially when you consider that reading a single byte in isolation is a very rare task, and it takes so long (compared to accessing a CPU cache) to find the data and transfer it that you might as well grab a whole load of bytes at the same time.

Byte-wise resolution in a HDD is like ordering rice by the grain in individual envelopes. Sure you could do it, but why the hell would you?

[maccam94]

I believe for performance and error correction features for spinning hard drives. Check out the tables and graphic in the overview section here[1]. Any write to a sector requires updating the checksum, which means the whole sector has to be read before the drive can complete the write. Since the sector is probably in the kernel disk cache already, the whole sector can be flushed to disk at once and the drive can compute the checksum on the fly (instead of flushing 1 byte, making the drive do a read of the sector, recompute the checksum, and then do a write).

1: https://en.wikipedia.org/wiki/Advanced_Format

[pkaye]

Because of the sector based error correction. I remember some earlier 8-bit disks has 128 byte sectors. Some newer flash SSDs have 1k-4k physical sector size for ecc coding efficiency so in those cases they will internally do a read-modify-write to handle a 512 byte logical sector size.

[tinus_hn]

Try designing the format keeing the required features in mind and you might seen why they have.

The hardware has to be able to read and write data from an arbitrary location and do some kind of checksum for error detection and they have to be able to efficiently transfer this data to the cpu.

[e12e]

.. And does it hold true for ram disks?

[avianlyric]

Yes it would, but you would be working in cache lines rather than disk blocks.

When working with RAM your PC will transfer a cache line of memory from RAM to your CPU caches. Again this is a feature of hardware limitations and trading off granularity against speed.

Your CPU operates many time faster than RAM so it makes sense to request multiple bytes at once, then your RAM can access those bytes, and line them up on it’s output pins ready for your CPU to come back and read them into cache. This give you better bandwidth. (It’s a little more complicated than this because the bytes are transferred in serial, rather than in parallel, but digging into the details here is tad tricky).

On the granularity point, the more granular your memory access pattern is, the more bits you need to address that memory. In RAM that either means more physical pins and motherboard traces, or more time to communicate that address. It also means more transistors are needed to hold that address in memory while you’re looking it up. All of those things mean more money.

And before we start looking at memory map units, that let you do magical things like map a 64 bit address space into only 16GB of physical RAM. Again the greater the granularity, the more transistors your MMU needs to store the mapping, and thus more cost.

So really the reason we don’t have bit level addressing granularity is ultimately cost. There’s no reason you could build a machine that did that, it would just cost a fortune and wouldn’t provide any practical benefits. So engineers made trade off to build something more useful.

[yencabulator]

For what it's worth, this is slowly starting to change. https://venturebeat.com/2019/09/20/optane-persistent-memory-...

tl;dr This new tech might do to SSDs what SSDs did to hard drives. Might.

[monadic2]

Well, one benefit would be a relaxation of constraints on filesystems. This might be worth performance loss.