Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
How much does available HD space affect drive speed?
- Thread starter eroxx
- Start date
- Sort by reaction score
As long as you leave around 10% or more space available for system files, caches, etc., the amount of storage you have used on your drive has no impact on performance. You won't gain any improvement by offloading anything from your drive.
If you're having performance issues, this may help:
I'll second that.
Normal HDD's show a decrease in speed as they fill up, but the fusion drive as implemented by Apple has a pretty constant speed up to 85% of the total capacity of both disks.
There is no question a regular HDD slows down as if fills up: http://macperformanceguide.com/Storage-WhyYouNeedMoreThanYouNeed.html
Normally I'd think the performance of a mostly-full Fusion Drive should be dominated by the HDD portion. However in this one test Mac Performance Guide did, performance stayed pretty flat up to the end (barring a big drop at the beginning): http://macperformanceguide.com/iMac-Fusion-fill-volume.html
However that was only one type of I/O -- apparently sequential. It cannot be fully indicative of overall FD performance across a range of free space. I have done many tests on my 3TB FD, and at 40% free space the QuickBench 20-100MB test still achieves about 670 MB/sec read, 310 MB/sec write, almost the same perf. as 93% free space. So in that case there was no big perf. drop as seen in the MPG test.
I may have posted before that FD slows down the same as a HDD when it fills up; if so that may not be correct. I'll do further testing on this and post my results.
However 800GB free on a 3TB FD is only about 25%, so just from a management standpoint it's time to start freeing stuff up. Otherwise you might run out of space at an inopportune time. Araxis Folder Size Explorer is a good tool to examine space consumption: https://itunes.apple.com/us/app/araxis-folder-size-explorer/id510309552?mt=12
This is the sort of message I like, with references and facts. You should post your complete results so everyone can partake.
OK I did some more testing and even down to about 1% free space, 3TB Fusion Drive performance is good -- at least on Black Magic and QuickBench large test. They were at least 90% as fast as on a nearly empty drive.
Those tests do a certain I/O size, profile, and "locality of reference". IOW they may re-read near the point they just wrote something -- just like a real app. Real apps don't always do brute force sequential I/O for hundreds of gigabytes. In those cases performance might fall off more, since FD's SSD cache could not help buffer it.
The bottom line is on these basic tests FD stays fast to the bitter end of capacity. To the extent that reflects real-world I/O activity, it means the old rule about maintaining free HDD space does not necessarily apply to FD.
This is different from what I've posted in the past, but I'd never actually tested it before now. So for eroxx, there's no immediate performance urgency to get more free space on his 3TB FD with 800GB free. However it's wise to periodically check your I/O performance with BlackMagic, QuickBench, or an application test.
Those tests do a certain I/O size, profile, and "locality of reference". IOW they may re-read near the point they just wrote something -- just like a real app. Real apps don't always do brute force sequential I/O for hundreds of gigabytes. In those cases performance might fall off more, since FD's SSD cache could not help buffer it.
The bottom line is on these basic tests FD stays fast to the bitter end of capacity. To the extent that reflects real-world I/O activity, it means the old rule about maintaining free HDD space does not necessarily apply to FD.
This is different from what I've posted in the past, but I'd never actually tested it before now. So for eroxx, there's no immediate performance urgency to get more free space on his 3TB FD with 800GB free. However it's wise to periodically check your I/O performance with BlackMagic, QuickBench, or an application test.