On October 23 along with the highly-anticipated announcement of the iPad 4, Apple rolled out new Macintosh computers that for the first time in an Apple product pairs an SSD with a conventional HDD to get the best combination of capacity, speed, and price. The company calls this its Fusion Drive, not to be confused with Fusion-io’s highly-regarded products.
The SSD Guy did not attend the announcement, and there is little on the Apple website. I contacted Apple, and they don’t have very much detail to share at this time. This is important to note, since several bloggers are currently posting their guesses of how the Fusion Drive works – these guesses must be taken with a grain of salt. The most informed appears to be on the Anandtech site.
Apple does explain that flash management is handled by the operating system. This approach is similar to that used by Intel for its Rapid Store Technology (RST), a critical part of the Ultrabook roll-out. RST, which is a part of the Intel-supplied Windows-compatible drivers, manages an SSD cache to accelerate an HDD in the Ultrabook.
The AnandTech review makes it sound as if the device uses a less-sophisticated rendition of tiering, in which data either resides in the SSD or in the HDD, but is never copied, with all software residing permanently within the SSD. This is a highly inefficient use of a costly resource.
In its promotional pieces Apple says: “Fusion Drive automatically and dynamically moves frequently used files to Flash storage for quicker access, while infrequently used items move to the hard disk.” This may only apply to data files, but not to software.
The Fusion Drive’s performance looks pretty good, and is illustrated in this post’s graphic, which compares the Fusion Drive’s performance in a Mac Mini against a Mac Mini using a standard 5,400 RPM HDD. A 4GB folder can be copied 3.7 times as fast (in other words it takes 27% as much time as with the standard HDD) and a photo can be imported 3.6 times as fast (or in 28% of the time that an HDD would take.) The system boot is twice as fast.
It seems odd that Apple would choose to use this unsophisticated approach which would force the design to pair a relatively large 128GB SSD with an HDD. One of the major benefits of caching (the approach used by Windows-based solutions and hybrid drives) is that you can get by with a very small (i.e. inexpensive) SSD, so that the price of the combined HDD-SSD system is relatively close to that of an HDD. At today’s prices Apple’s 128GB SSD is likely to cost roughly twice as much as the terabyte HDD it’s teamed with. Compare this to Intel’s RST that gets by with a minimum of 18.6GB of SSD, or NVELO’s 16GB minimum for its Dataplex. The recently-announced hybrid drives from Seagate and WD sport 16GB of NAND flash as well, Toshiba’s has 8GB, and Seagate’s popular original Momentus XT hybrid which has been shipping since mid-2010 performs well with only 8GB.
According to Apple: “Fusion Drive is an innovative new storage option that gives customers the performance of flash storage and the capacity of a hard drive. It combines 128GB of flash with a standard 1TB or 3TB hard drive to create a single storage volume that intelligently manages files to optimize read and write performance. Fusion Drive adapts to the way you use your iMac and automatically moves the files and apps you use most often to flash storage to enable faster performance and quicker access.”
It’s unfortunate for everyone that Apple chose to use the name “Fusion” for its fast storage. This will doubtlessly lead to confusion about whether or not Fusion-io played a role in this system.
3 thoughts on “Apple’s Fusion Drive – An SSD Cache for the Macintosh”
In terms of power consumption, speed, capacity, effectiveness of cache algorithms and degradation over time, which do you think is better? Using a hybrid hdd or a drive with an SSD cache?
Ben, This is a really tough question to answer.
Some caches use a copy-back or a write-through policy, either of which increases flash writes making a larger cache desirable, or perhaps even SLC. Others preserve the life of the NAND by using a read-only approach. This will have a lower hit rate (although how low it is depends on the application) increasing power consumption and providing slower performance. So you’re trading off performance and power consumption against cost and lifetime.
I don’t think there’s a hard-and-fast rule that I can give you for hybrids vs. external caches, but I believe that the Apple approach, which doesn’t move code around, is an expensive way to go that probably results in very low wear, but may not provide significantly better performance than a much smaller SSD managed with a more sophisticated algorithm.
Hope that helps.
Comments are closed.