4x NVME to PCIE with PLX

[happy_home]

@zdude Have you made any further progress with the configuration options? Thank you for all your effort so far.

 

[richard.dzavoronok]

For those in Germany:

Viking U20040-02 2.5" U.2 NVMe SSD Module | eBay.de

Mit seiner fortschrittlichen Architektur und umfangreichen Spezifikationen bietet es herausragende Leistung und Zuverlässigkeit. Es ist sofort einsatzbereit und bietet eine zuverlässige Speicherlösung für anspruchsvolle Anwendungen.

www.ebay.de

It's quite cheaper in their own shop Viking U20040-02 2.5" U.2 NVMe SSD Module

€19,99 + €13,90 shipping within europe (flat).
48 pieces in stock as now.

 

[happy_home]

I bought a couple of them and they threw no PCIe errors compared to the PCIe switch card I tested (delock 90504). Each one drew about 6.9 W at the outlet. As mentioned before, they get uncomfortably hot during idle load. I’m wondering if it would make sense to attach two 2.5-inch copper heatsinks—one on each side—using thermal adhesive to provide suffient cooling with moderate airflow, or if this direct heat transfer to the NVME drives makes things worse.

Persistence 100X70MM Reines Kupfer KüHler KüHlköRper Flossen PC Htpc Laptop Notebook Mechanische SSD Festplatte Router KüHler: Amazon.de: Computer & Zubehör

Persistence 100X70MM Reines Kupfer KüHler KüHlköRper Flossen PC Htpc Laptop Notebook Mechanische SSD Festplatte Router KüHler: Amazon.de: Computer & Zubehör

www.amazon.de

 

[zdude]

@zdude Have you made any further progress with the configuration options? Thank you for all your effort so far.

Unfortunately life got in the way of the home lab over the past few months and not a lot has happened (my main ceph cluster has a node down that I haven’t brought back online yet).

I got to the point that I believe I was able to find most of the structure in the binary from the switches themselves, it was massively helpful that the servers I am using for this testing have a different model PLX from the same generation to compare to.

I have been unable to modify then flash the modified configs, I think I need to look for and find a checksum somewhere but have not had the time. If people are interested I can find the python code I was using to extract values/parse what looked like arrays from the raw binary files and post it on GitHub. Honestly getting some formal documentation would be the best path forward here.

 

[celemine1gig]

My guess is, that you would need those Microchip specific tools for the FW. Otherwise all bets are off, concerning the question of how long it will take to reverse-engineer this. Can go fast, can take ages. So, either we'll get the tools, or the idea of reconfiguring seems kind of unrealistic.

I definitely know, that the challenge can be very interesting, but then you only have that much time available, as you already showcased.

Edit:
Seems like the "ChipLink" utility is what we would require.
See here: https://support.microchip.com/s/article/Accessing-ChipLink-Tool

However, of course that is under NDA.