still waiting to connect my cd-burner over NVME. why arent there any forward looking standards?
Linux 6.13 Rolling Out NVMe 2.1 Support & NVMe Rotational Media
Collapse
X
-
Originally posted by davidbepo View Postalso wake me up when an HDD comes close to even touching sata limits, this is just nonsense
When you're rolling out massive drive arrays for S3 storage, ZFS nearline arrays, Ceph clusters, etc, then NVME has a lot of benefits over legacy SATA/SAS in both simplifying the connection points, as well as removing some upper limits to how many devices can be attached without needing more controllers.
Similarly, hanging lots of rotational disks off a single 6Gbit controller with port multipliers is absolutely a bottleneck. Again, see the video for how simplified PCIE switching and NVME results in simpler hardware and higher speeds.
The video also talks about what it looks like when every bit of compute and IO is all on the same fabric. For future workloads, having multiple classes of storage, network, GPUs and the like all on the same PCIE/NVME fabric simplifies a lot of problems we have in high end clustering.
Lots of applications for this beyond what a single drive looks like in 2024. NVME looked pretty silly even for flash when it first arrived, because we couldn't hit those speeds back then. But it was clear that it was a necessary change as things moved forward. Limiting things to today's technology is not how the industry works.
Originally posted by davidbepo View PostNVMe requires a new physical connector and form factor over SATA/SASLast edited by elvis; 18 November 2024, 06:56 PM.
Comment
-
-
Originally posted by Joe2021 View PostSounds great. Nevertheless it is so sad that most consumer SSDs only support exactly one namespace at once. That is as useful as a partition table scheme capable to hold one partition max - aka pointless!
Is there something I'm missing? Are namespaces for VMs or containers? You might as well run some complicated LVM setup instead and do it in software.Last edited by ahrs; 18 November 2024, 08:18 PM.
Comment
-
-
support for offloading some host processing to NVMe storage devices
cant be bothered to check, but id be surprised if the infrastructure (or even the blocklist itself) arent already created.
Comment
-
-
Originally posted by caligula View Post
Moore's law. SATA3 is limited to 600 MB/s per drive. NVMe already supports PCIe 5.0 16x drives already transfer up to 63 GB/s and more is expected with PCIe 6.0. You just need to spin the motor a bit faster.
I suspect it's about cutting a smaller disk to fit on the M.2 form factor. With the evolving magnetic density, you can still have a very high transfer rate and over 1TB capacities. A smaller radius disk reduces seek times, which was already exercised long time ago. Look up "short stroking hard disk".
Comment
-
-
Originally posted by caligula View Post
Moore's law. SATA3 is limited to 600 MB/s per drive. NVMe already supports PCIe 5.0 16x drives already transfer up to 63 GB/s and more is expected with PCIe 6.0. You just need to spin the motor a bit faster.
Comment
-
-
Originally posted by davidbepo View Postman people here with the memes and missing the point so hard
NVMe requires a new physical connector and form factor over SATA/SAS, also wake me up when an HDD comes close to even touching sata limits, this is just nonsense
Comment
-
-
Originally posted by dlq84 View Post
Why not? NVME support vastly more command queues and is both faster and more efficient over fiber channel. Just to name a couple of reasons.
Comment
-
Comment