No, your next MacBook's storage won't be 1,000x faster

Not even close. Why? NVM totally upsets existing storage stacks - and the Mac OS stack is creakier than most. Here's why.

Intel is promising availability of 3D Xpoint non-volatile memory (NVM) this year, at least in their Optane SSDs. That has led some to speculate that MacBooks, most of which already use the super fast NVMe interface, will jump to Optane SSDs in the next year.

Here's why it won't happen

special feature

The Evolution of Enterprise Storage

How to plan, manage, and optimize enterprise storage to keep up with the data deluge.

Read More

Speed changes everything. I/O today is interrupt based. An I/O request is issued, CPU state is saved, other tasks are handled and upon I/O completion, an interrupt tells the CPU to swap in the process that issued the I/O and continue working. That works today because a CPU can perform a couple of million operations during the millisecond it takes to read an SSD.

However, with storage that is 1,000x faster, the overhead of process swapping becomes huge relative to the much faster I/O. Which means either you change how you handle I/O - using polling instead of interrupts - or you sacrifice a major portion of the speed advantages of 3D Xpoint storage.

After a brief flirtation with the 21st century ZFS file system, the Mac OS file system team has continued to patch HFS+, a relic of 80s, and whose design bears a strong resemblence to the RT-11 file system from 1970. Unlike Microsoft, who has done much to modernize NTFS under the hood, Apple has been content to bake HFS+ into ever more services, making updating the Mac storage stack even more difficult.

Addressability. 3D Xpoint is byte addressable, like DRAM, while NAND flash is only block writable and page readable. That is a major advantage over flash, but only if your file system can take advantage of it.

For example, byte addressability means that data can be updated in place. Today's file systems assume 4k blocks so the bandwidth advantages of byte addressing are overwhelmed by file system assumptions. Again, HFS+ is ill-suited to use this capability.

Price. Intel/Micron has been most reticent on 3D Xpoint pricing, saying only that it will come in between flash - ≈20¢/GB - and DRAM - ≈$5/GB. That's quite a range.

Given that Intel has been facing margin pressure, and that 3D Xpoint is SLC-only initially, I'm forecasting that Optane SSDs will come in at $2/GB to pay Xpoint startup costs and make Intel some bank. $2/GB doesn't sound like much more than what Apple charges for MacBook flash - around $1/GB - but Apple also wants 40+ percent margins on Macs, which means pricing Optane SSDs at ≈$3.33/GB.

Even diehard Macophiles will have difficulty paying a 3x price for performance that won't be that much better than current SSDs.

The Storage Bits take

Good on Intel's PR group for getting PC pubs to push the 1,000x faster theme. But bad on those pubs for believing it.

It is fiendishly difficult to successfully bring a new storage medium to market. It took Toshiba 25 years to get flash to the point where it took off in a big way.

Intel's hyping of 3D Xpoint is understandable. They have to economically ramp up production, while also asking for major changes to how systems use memory and/or storage to take advantage of the new technology.

This is one case where I'd expect Windows 10 and the revamped NTFS file system to take advantage of Optane SSDs sooner than the Mac OS. NTFS is a much more modern file system and the MS engineering team has already incorporated major improvements under the NTFS API.

Yes, 3D Xpoint will be faster. But not 1,000x faster, until the rest of the storage stack is adapted to use the new and different technology.

Comments welcome, as always. Please Apple, surprise us with a new storage stack that can take advantage of your latest hardware!

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All