3 ms·
> But Core Storage, the technology responsible for keeping the most frequently accessed files on the SSD, can only see data on your drive at the block level, no
by kccqzy 8y ago
> But Core Storage, the technology responsible for keeping the most frequently accessed files on the SSD, can only see data on your drive at the block level, not the file level. It can’t tell the difference between an app that needs to be launched quickly or a document that doesn’t need a lot of extra speed; all it can see is how frequently blocks are accessed. APFS changes that. Fusion Drives can now move files to the SSD based not just on how frequently they’re accessed, but also based on how much the type of file will benefit from an SSD. APFS will store all file metadata on your SSD, too, speeding up Spotlight searches and metadata lookups in the Get Info window.
Can someone more knowledgeable than me explain how in the world this is an improvement? The old Core Storage model perfectly makes sense to me: it strictly operates beneath the level of a file system, combining multiple physical volumes into a logical volume family which contains a logical volume. It moves blocks between SSD and HDD using frequency of access. Now, why should we suddenly need to look at the type of file to determine which block device to store the data? Even if it's an application, but if it's not frequently used, why should we keep it on SSD?
And filesystem metadata should be even more obvious. Since basically all file operations need the directory entries and the inode information, the old Core Storage-based system will automatically move them to the SSD. Why do we need to explicitly tell the filesystem to do this?
Overall this feels like a step backward to me. I like systems that are dynamic and self-adjusting, not those with hard-coded rules and heuristics.
Also I'm a bit disappointed as the author, while producing a fine review to read, didn't quite do the deep-dive I had expected for a system that's arguably the most important in an OS—a system in charge of storing user data.
- monocasa 8y agoJust in general, when optimizing for different latency use cases, looking at averages falls flat in a lot of cases. For instance you probably want fast startup, but how often does the startup code get executed?
- coldtea 8y ago>Now, why should we suddenly need to look at the type of file to determine which block device to store the data? We don't "need" to, it's an extra optimization. >Even if it's an application, but if it's not frequently used, why should we keep it on SSD? Because if we have the space to spare on the SSD, we'd appreciate the faster launch when we do try to use that app. And for other types of files it can now decide to move them or not move them based on the benefit from the faster load, not just the frequency of access (and thus optimize the SSD use).
- pertymcpert 8y agoMaybe because performance analysis showed that it was better to do it the new way.