NVME - Hardware Modification for Battery/BBU/CAP - Custom

[webdawg]

So I know this is crazy, but ever since I started to think about it....

I know the intel pci-e enterprise nvme solutions, and liquid solutions have power solutions that keep the device on after shutoff, and that this is needed for a proper ZIL device.

Any hacks around this?

I mean can I throw a few caps on this, or batteries(s), and have my solution: https://www.amazon.com/ADWITS-Adapter-Controller-Supports-Bifurcation/dp/B08348376V/

Solder them? Or pin, and cable them?

Has anyone built anything like this, can you link it?

I see this stuff too:

PB-2500J SuperCap UPS, Industrial Power Backup Module - Neousys Technology

Neousys PB-2500J SuperCap UPS is an innovative wide-temp power backup module for demanding industrial applications utilizing supercapacitor technology

www.neousys-tech.com

Supermicro Mounting Kit for Flash or Battery Backup Module (PCIe Slot Kit) | AVADirect

Supermicro Mounting Kit for Flash or Battery Backup Module LSI BKT-BBU-BRACKET-05 Remote Mounting Board

www.avadirect.com

Are these guys just strapping capicitors to these things, and calling it a day:

View attachment 45547

Same with intel?

Do they flush the nvme buffer, when their is a power loss?

What is liquid talking about here too:

View attachment 45549

ECC? What? Like checksums, like what the rest of what hard drives already do?

 

[HoneyBadger]

Unfortunately it won't do what you're asking. SLOG performance isn't about the presence of an inline capacitor or battery power pack, but rather that the drive knows it has one and can respond in a fast manner to the "flush your cache" request.

A drive that knows it has battery backup can take a "lazy" approach to a cache flush - its RAM buffer is protected from power loss, so it can respond back to the HBA/initiator with "Okay, all of my volatile cache has been made safe" and then lazily commit that data from RAM to physical NAND knowing that it has enough internally stored power to complete that even if the power is cut immediately after that "flush" command is received.

A drive without that knowledge in its firmware - regardless of the power backup status - will have to reply "hold on" while it commits its RAM to physical NAND, and only after that can it say "data in volatile cache has been made safe"

The mounting kit for FBWC/BBU's is just that - a physical carrier card only. There still has to be a storage device (SSD, NVRAM PCIe card, NVDIMM) with firmware that can handle the cache flush command appropriately.

Side note - the PCIe carrier card you linked to, I wouldn't suggest it for SLOG. It has a PCIe switch on it, which will add latency, and the name of the game with SLOG is to minimize that as much as possible. Go for a direct-attached card and use a motherboard with bifurcation support if you require multiple physical SSDs in a single slot.

 

[webdawg]

Thanks!