TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Kioxia Demonstrates Raid Offload Scheme for NVMe Drives

36 pointsby vanburen9 months ago

7 comments

sliken9 months ago
Seems like quite a bit of complexity for a dubious win. Today&#x27;s CPUs are REALLY fast, even a single core of 64 or more cores that are common on servers today:<p><pre><code> [ 478.047970] raid6: avx2x2 gen() 60473 MB&#x2F;s [ 478.115971] raid6: avx512x1 gen() 53469 MB&#x2F;s [ 478.149971] raid6: avx512x2 gen() 57067 MB&#x2F;s </code></pre> Especially since the data is coming from the CPU anyways, so likely caches are warm. It also means that node you have to send a stripe of data to a single NVME, which likely has much less than 60GB&#x2F;sec of checksum speed, then initiate transfers to every other drive in the stripe. Not to mention the NVME drive likely doesn&#x27;t have ECC memory and any resulting memory errors are unlikely to be visible to the OS.<p>Just seems like hardware RAID with all the same problems, likely not as fast as software RAID, harder to manage, a unique set of tools per vendor, harder to have global spares, and doesn&#x27;t work with filesystems that do their own redundancy like ZFS.
评论 #41258720 未加载
评论 #41259023 未加载
评论 #41259357 未加载
mrktf9 months ago
I imagine these kind of schemes can be implemented as sort of on device eBPF filter (in layman terms CUDA, but for storage). It would allow deeper integration with system for example have hardware accelerated&#x2F;integrated lvm (obviously speed would depend on use case, less win for thin volumes, more advantages for raid and so on). Or from other side have deeper integration with filesystems such as zfs, btrfs, bcachefs.
评论 #41258795 未加载
评论 #41260698 未加载
评论 #41258510 未加载
przemub9 months ago
When I first bought Kioxia flash I thought it&#x27;s a random Chinese knockoff. Shame they ditched Toshiba brand on these.
baruch9 months ago
NVMe drives fail at a fairly low rate so this is an optimization for a very small edge case and since they are also very fast it&#x27;s not like you&#x27;ll be doing a rebuild for 6+ hours like with HDDs.<p>It also doesn&#x27;t change anything for distributed storage.
评论 #41260542 未加载
评论 #41260413 未加载
评论 #41260382 未加载
ein0p9 months ago
I’d much prefer NVMe colocated compute. Imagine a columnar storage engine able to filter and aggregate data during scans without reading it through PCIe, for example.
评论 #41259588 未加载
jamesfmilne9 months ago
HW RAID is dead, they need to get over it.<p>We&#x27;ve had good experience with Xinnor, but it&#x27;s a shame it&#x27;s proprietary.<p>I&#x27;d love to see a high-performance open-source erasure coding solution for NVMe. The built in offerings in Linux are not cutting it.
znpy9 months ago
I wonder how (if?) this will interact and integrate with the current software stack and the various volume-managing filesystems (zfs but also btrfs).
评论 #41259174 未加载