Open-Channel SSD

[zir_blazer]

Just read this on Phoronix, a Linux News site:

http://www.phoronix.com/scan.php?page=news_item&px=Open-Channel-SSD-Linux-V4

http://events.linuxfoundation.org/sites/events/files/slides/LightNVM-Vault2015.pdf

Basically, there is an implementation of a Driver in the Linux Kernel that is intended to allow for a new type of SSDs with no FTL (Flash Translation Layer) controller, where NAND ICs are directly accessed. They should put load on the computer, but seems like they could be slighty more performing, and cheaper. Looks really interesing to me.

 

[ksec]

Well the best thing about this is Software Defined SSD. No Garbage Collection or all sort of wired thing and optimization happening inside controller. As i see Controller getting much more complex and the hidden complexity of it will one day bite back.

 

[TheUAoB]

Just read this on Phoronix, a Linux News site:

http://www.phoronix.com/scan.php?page=news_item&px=Open-Channel-SSD-Linux-V4

http://events.linuxfoundation.org/sites/events/files/slides/LightNVM-Vault2015.pdf

Basically, there is an implementation of a Driver in the Linux Kernel that is intended to allow for a new type of SSDs with no FTL (Flash Translation Layer) controller, where NAND ICs are directly accessed. They should put load on the computer, but seems like they could be slighty more performing, and cheaper. Looks really interesing to me.

From the results posted on LKML, the performance (latency) is *much* faster, with only a small host side overhead.

 

[AlienTech]

Many companies are using such drives and writing their own firmware to speed up access times. Especially using raid stripping and such to get incrediable read/wrie speeds. I saw baidu was using thousands of such ssd's to do search caching.. I bet everyone else is doing it as well. Using their own software speeds up things many times over.

 

[zir_blazer]

Been just reading this News article. There is another Thread going on about DRAM-less SSD Controllers, but point is that if the new NVMe 1.2 spec allows for reserving host RAM to use as Cache for the SSD, you're halfway over what Open-Channel is supposed to do. If a next NVMe specifications allows for direct NAND access bypassing the SSD Controller while host reserves RAM for Cache, voila, Open-Channel SSD equivalent in the standard NVMe specification.

 

[cbn]

Just read this on Phoronix, a Linux News site:

http://www.phoronix.com/scan.php?page=news_item&px=Open-Channel-SSD-Linux-V4

http://events.linuxfoundation.org/sites/events/files/slides/LightNVM-Vault2015.pdf

Basically, there is an implementation of a Driver in the Linux Kernel that is intended to allow for a new type of SSDs with no FTL (Flash Translation Layer) controller, where NAND ICs are directly accessed. They should put load on the computer, but seems like they could be slighty more performing, and cheaper. Looks really interesing to me.

Embedded FTLs: No Future

• Dealing with flash chip constraints is a necessity
– No way around some form of FTL

• Embedded FTLs were great to guarantee
adoption, but have critical limitations:

–Hardwire design decisions about data placement,
overprovisioning, scheduling, garbage collection and
wear leveling
–Based on more or less explicit assumptions about the
application workload
–Resulting in redundancies, missed optimizations and
underutilization of resources

Market Specific FTLs

SSDs on the market with embedded FTLs
targeted at specific workloads (90% reads) or
applications (SQL Server, KV store)

•FTL is no longer in the way of a given
application

•What if the workload or application changes?

•What about the other workloads or
applications?

Assuming NVMe adopts this, I wonder what the power consumption differences would be for a Windows laptop using classic client applications?

Are Windows applications still an instance where having a SSD controller is still more efficient? (ie, dedicated hardware vs. using general purpose CPU cores)

With that mentioned, assuming an SSD controller (either separate or integrated into the SoC) were more efficient for typical Windows client applications, I could still see this Open Channel SSD taking off for desktop (and maybe some specialty forms of laptops).

 

[VirtualLarry]

For your average client PC, if it were experiencing instabilities, I would not want to use a software-controlled SSD. There are things that a discrete controller-based SSD can do to both do background tasks, and cleanup in case of host system or power failure.

I don't think that I would trust a controller-less SSD to perform those tasks as well.

Remember "WinModems"? And how they sucked, sucking down CPU time, less stable connections, general driver incompatibilities / instabilities? Would your average user want to go down that road again?

Edit: To say nothing about making your SSD dependent on a particular OS and even version? Could wreak havoc with Windows / Linux dual-boots.

 

[cbn]

Edit: To say nothing about making your SSD dependent on a particular OS and even version? Could wreak havoc with Windows / Linux dual-boots.

For booting, I'm imagining SoCs will eventually have an integrated SSD controller (PCIe 3.0 x 4, etc).

This iSSD controller (cooled by the CPU HSF) would control one pool of NAND (composed of X number of packages, each with a certain number of die in them).

EDIT: Then after that happens, maybe we see folks experimenting with additional M.2 form factor devices for desktop/workstation/Server (some with controllers and NAND, but maybe some with just NAND?)

 

[cbn]

Open Channel prototype by Lite-ON:

http://www.thessdreview.com/flash-m...rotoype-u-2-ssd-and-unveils-ep3-m-2-nvme-ssd/