3 ms·

Tags are fun when you have a few thousand items to test your MVP with. It gets much less fun when you have millions of items with thousands of tags, all on a fl

by kijin 9y ago

Tags are fun when you have a few thousand items to test your MVP with. It gets much less fun when you have millions of items with thousands of tags, all on a flat hierarchy. On the other hand, when you're stuck with a flat hierarchy anyway (e.g. thousands of pictures, all named DCIMxxxx.jpg), tags can be more useful. But only if they're automatic. I want the best of both worlds. I want to organize my stuff into folders and use tags to search for individual items. There's no need to be a purist on either side. "Designing better file organization around tags" is a good thing. "Designing better file organization around tags, not hierarchies" is not.

nayuki 9y ago
I understand your concerns, and they are indeed valid. First off, I doubt that managing millions of files in a traditional hierarchical file system is fun either. You'd likely run into problems with making unique names, sharding folders, and categorizing files that logically belong in multiple places. I also have some worries that existing file systems (say NTFS or XFS) will behave or perform well with millions of files. I believe that implementing tags is a starting point for the problem of managing millions of files in a sensible way. Speaking of thousand of pictures, what I really want is to dump all my photos into one folder. Right now, tools are ill-equipped to deal with large folders, so I am forced to manually create a new folder for every thousand or so items.
- cyphar 9y ago
  > I also have some worries that existing file systems (say NTFS or XFS) will behave or perform well with millions of files. I believe the mantra for XFS is "if you have large or lots, use XFS". XFS has a lot of optimisations for metadata operations which should mean it's better than most filesystems for lots-of-files and large-files cases (Dave Chinner has given several talks about the performance characteristics of XFS with "large or lots" cases).
enobrev 9y ago
I absolutely agree with this, except to add that it seems, in theory, in the case of a 100% tag-based file-system, there would never (*very rarely) be a flat list where you have to scroll through millions of files. The UX of a single flat list with millions of files named DCIMxxxx.jpg is a limitation of the current format and doesn't make sense when so much information can be generated about our files upon creation. In this hypothetical FS, all files would have dynamically generated tags for created date/time, modified date/time, access date/time, originating application, given filename, owner, group, permissions, format, geo-data if available, originating hostname, EXIF data should all be first-class tags as prominent as the others, and so on. I visualize it to work, by design, something like how google photos UX works, grouping everything by EXIF data automatically before you ever start organizing things on your own. Just as well, with any files created manually, in the application "tagging" should be just as prominent a function as "naming" is right now.
- TeMPOraL 9y ago
  Part of the adoption problem here would be trust. Hierarchical filesystems are something we're used to, and we can trust they're implemented correctly. That means, if I visit a folder and see some files, I know what I see is all of the files there (+/- hidden file settings); if something is missing, it's not there, period. Tag search is a search. Can be broken. Can be optimized in a way that causes it to lie. I look at the results, and I'm not sure if they're complete. Maybe the file I'm looking for is really not there, or maybe the search gave up too early. Or the tag was slightly misformatted? Maybe I'm too used to the old thing, but I like the notion that there's one canonical tree structure that makes all my data reachable. In case I've misplaced something, the search space of all paths through the filesystem tree (or a subtree of interest) is vastly smaller than the search space of all possible values of all relevant tags.
  - enobrev 9y ago
    You make excellent points, although I think I the issues you raise exist in our current filesystems as well. Provided the FS is indexed properly, opening a tag should show all files associated with that tag immediately. Just like opening (or listing) a directory does. And in the case of search, it absolutely sucks on today's filesystems. Don't get me wrong, find and grep are incredible tools. I simply mean that it's not like searching for files beyond the hierarchy is known for the pleasurable UX. The only way I know a grep of a whole drive or deep directory is done is because I get my blinking cursor back. At the very least with a proper tagging system, we would be inherently familiar with the indexes available to us.
    - TeMPOraL 9y ago
      > And in the case of search, it absolutely sucks on today's filesystems. It does. You mention grep, I'd even mention find - half of the time I'm wondering whether it has searched everything I wanted, or I misspelled the command. Or file search in Windows (Vista+) - I just don't trust it; I'm pretty sure it missed some data in the past for one reason or another. Now with traditional file systems, I at least have the file tree. With tag-based systems, I'd only have search - so it better be trustworthy, both in reality and UX-wise. It needs to project the feeling of correctness and completeness or results. > The only way I know a grep of a whole drive or deep directory is done is because I get my blinking cursor back. The only way I know a find of a whole drive does what it's supposed to be doing is because it emits a stream of "find: `/some/path': Permission denied" messages. > At the very least with a proper tagging system, we would be inherently familiar with the indexes available to us. Fair enough.

by kijin 9y ago

nayuki 9y ago

I understand your concerns, and they are indeed valid. First off, I doubt that managing millions of files in a traditional hierarchical file system is fun either. You'd likely run into problems with making unique names, sharding folders, and categorizing files that logically belong in multiple places. I also have some worries that existing file systems (say NTFS or XFS) will behave or perform well with millions of files. I believe that implementing tags is a starting point for the problem of managing millions of files in a sensible way. Speaking of thousand of pictures, what I really want is to dump all my photos into one folder. Right now, tools are ill-equipped to deal with large folders, so I am forced to manually create a new folder for every thousand or so items.

cyphar 9y ago
> I also have some worries that existing file systems (say NTFS or XFS) will behave or perform well with millions of files. I believe the mantra for XFS is "if you have large or lots, use XFS". XFS has a lot of optimisations for metadata operations which should mean it's better than most filesystems for lots-of-files and large-files cases (Dave Chinner has given several talks about the performance characteristics of XFS with "large or lots" cases).

enobrev 9y ago

I absolutely agree with this, except to add that it seems, in theory, in the case of a 100% tag-based file-system, there would never (*very rarely) be a flat list where you have to scroll through millions of files. The UX of a single flat list with millions of files named DCIMxxxx.jpg is a limitation of the current format and doesn't make sense when so much information can be generated about our files upon creation. In this hypothetical FS, all files would have dynamically generated tags for created date/time, modified date/time, access date/time, originating application, given filename, owner, group, permissions, format, geo-data if available, originating hostname, EXIF data should all be first-class tags as prominent as the others, and so on. I visualize it to work, by design, something like how google photos UX works, grouping everything by EXIF data automatically before you ever start organizing things on your own. Just as well, with any files created manually, in the application "tagging" should be just as prominent a function as "naming" is right now.

TeMPOraL 9y ago
Part of the adoption problem here would be trust. Hierarchical filesystems are something we're used to, and we can trust they're implemented correctly. That means, if I visit a folder and see some files, I know what I see is all of the files there (+/- hidden file settings); if something is missing, it's not there, period. Tag search is a search. Can be broken. Can be optimized in a way that causes it to lie. I look at the results, and I'm not sure if they're complete. Maybe the file I'm looking for is really not there, or maybe the search gave up too early. Or the tag was slightly misformatted? Maybe I'm too used to the old thing, but I like the notion that there's one canonical tree structure that makes all my data reachable. In case I've misplaced something, the search space of all paths through the filesystem tree (or a subtree of interest) is vastly smaller than the search space of all possible values of all relevant tags.
- enobrev 9y ago
  You make excellent points, although I think I the issues you raise exist in our current filesystems as well. Provided the FS is indexed properly, opening a tag should show all files associated with that tag immediately. Just like opening (or listing) a directory does. And in the case of search, it absolutely sucks on today's filesystems. Don't get me wrong, find and grep are incredible tools. I simply mean that it's not like searching for files beyond the hierarchy is known for the pleasurable UX. The only way I know a grep of a whole drive or deep directory is done is because I get my blinking cursor back. At the very least with a proper tagging system, we would be inherently familiar with the indexes available to us.
  - TeMPOraL 9y ago
    > And in the case of search, it absolutely sucks on today's filesystems. It does. You mention grep, I'd even mention find - half of the time I'm wondering whether it has searched everything I wanted, or I misspelled the command. Or file search in Windows (Vista+) - I just don't trust it; I'm pretty sure it missed some data in the past for one reason or another. Now with traditional file systems, I at least have the file tree. With tag-based systems, I'd only have search - so it better be trustworthy, both in reality and UX-wise. It needs to project the feeling of correctness and completeness or results. > The only way I know a grep of a whole drive or deep directory is done is because I get my blinking cursor back. The only way I know a find of a whole drive does what it's supposed to be doing is because it emits a stream of "find: `/some/path': Permission denied" messages. > At the very least with a proper tagging system, we would be inherently familiar with the indexes available to us. Fair enough.