add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube

Store MI & Seagate Momentus XT

bobbarley
  • 18 months ago

Has anyone ever tried using Store AI to accelerate a hybrid hard drive?

I have a Seagate Momentus XT lying around doing nothing, and am building a Ryzen based system this weekend. I thought it might be fun to try it and see what happens.

So essentially a 500GB hard drive with a 4GB NAND chip that caches regularly used files, connected to an SSD via Store MI, that also cashes regularly used files!

I envisage it will either allow time travel, or burn my house down :D

Comments

  • 18 months ago
  • 1 point

It wouldn't surprise me if you lost performance/crash because the system was trying to cache and retrieve the files from two places at the same time.

  • 18 months ago
  • 1 point

Seeing as how most disk drives already have a cache, although usually smaller and volatile, I'd expect it to work just fine.

The drive cache isn't exposed to the OS so StoreMI has no idea that there's anything different. It might do some unnecessary shuffling to bring some commonly used files onto the SSD when it doesn't strictly need to, but otherwise it should all work, and hard drive access might be unexpectedly quick if they hit the nvcache part of the drive.

By the way, I understand that StoreMI is a tiering solution, not a caching solution. Tiering places commonly used files off the hard disk onto the SSD, caching puts a COPY of commonly used files or file blocks on the SSD. Both have plusses and minuses.

[comment deleted]
  • 18 months ago
  • 1 point

Sounds interesting, please report the results

  • 18 months ago
  • 1 point

I will try to do some scientific tests if possible!

  • 18 months ago
  • 1 point

I'd assume that results would be the same as for StoreMI and a drive with only a DRAM cache (measured in MB). One difference is that as a tiering system, StoreMI will want to move the data to the SSD, so any data previously accessed on the HDD will now be on the SDD (completely wasting the cache's main strength). You might still get a lot of hits on the HDD "reading ahead", but you would get most of them with a tiny DRAM buffer (typically) unless there were just too many files open to buffer.

Sort

add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube