Apple's SATA controller underperforms with SSD
Not even Snow Leopard will fix the 10MBps hit OS X takes with SSD performance
That kind of sucks. I kind of wish Apple would get back to writing software that performs well rather than just looks good.
Quote:
The only way to avoid this is by buying an external RAID controller that comes with its own drivers, independent of the built-in SATA drivers. Now, brace for impact - upcoming Mac OSX Snow Leopard WILL NOT fix this one in its initial release, we will have to wait for an Apple Update, if it ever comes out. The issue is present in all Mac OS X releases with SATA drive support, so you lose 10MB/s if you use a very fast SSD drive.
The only way to avoid this is by buying an external RAID controller that comes with its own drivers, independent of the built-in SATA drivers. Now, brace for impact - upcoming Mac OSX Snow Leopard WILL NOT fix this one in its initial release, we will have to wait for an Apple Update, if it ever comes out. The issue is present in all Mac OS X releases with SATA drive support, so you lose 10MB/s if you use a very fast SSD drive.
That kind of sucks. I kind of wish Apple would get back to writing software that performs well rather than just looks good.
Comments
Not even Snow Leopard will fix the 10MBps hit OS X takes with SSD performance
That kind of sucks. I kind of wish Apple would get back to writing software that performs well rather than just looks good.
Too bad the article is light on specifics and doesn't discuss whether it's a controller bottleneck, a design bottleneck in architecture I/O, a dtrace showing it's an I/O device driver error in their code, or a combination.
More importantly, this "up to" burst speed is spurious. I want to know sustainable speeds, nominal transfer rates, etc. I would love to test actual performance rates over heavy loading to see how they stack up.
My guess would be Dean Reece's team have written in a compromise to hit that sweet spot, until unreported issues are resolved [on Apple's end, Intel's end or both].
Too bad the article is light on specifics and doesn't discuss whether it's a controller bottleneck, a design bottleneck in architecture I/O, a dtrace showing it's an I/O device driver error in their code, or a combination.
More importantly, this "up to" burst speed is spurious. I want to know sustainable speeds, nominal transfer rates, etc. I would love to test actual performance rates over heavy loading to see how they stack up.
My guess would be Dean Reece's team have written in a compromise to hit that sweet spot, until unreported issues are resolved [on Apple's end, Intel's end or both].
I know I was far too harsh on Apple. I read the article again and it just didn't make sense.
Even if the SATA controller doesn't improve it's only losing 10MBps on sequential read/write. Many people would rather have gonzo random read/write and therefore would like the Intel SSD.
I'd say that it would likely make more sense to re-arch not only the SATA controller but implement other functionality that optimizes for SSD. That's probably what Apple's going to do. SSD isn't mainstream yet but in a few years it will be and we'll have SATA 6Gbps and more knowledge of how to optimize for SSD.
I hope OCZ and Apple work out the issues with using SSDs better on Apple Macs.
I use a SSD daily on my Mac Mini. I really like it. I have a OCZ Solid with a JMICRON controller (I know...) but I just use it as an external USB drive for storing my essential data files on the Mac Mini, and not as a Bootcamp or System drive.
I hope OCZ and Apple work out the issues with using SSDs better on Apple Macs.
I'm sure they will and frankly I think that Apple will do more than simply update their SATA driver. Optimizing for SSD requires more work and I expect that like Microsoft, Apple is working on changing more than a few things in their storage subsystems to accomodate the peculiarities of NAND (like not having to account for rotational latency)
I think Apple deals with issues based on the magnitude of the group it affects. This issue probably affects about 20 people right now. Once the 64-bit driver switch happens and everything gets stable again, they can plan for these fixes. I imagine there will be a few significant maintenance updates to Snow leopard.