SOLVED Chelsio T580-LP-CR Windows-TrueNAS DAC experiences?

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
Hello all:

I'm presently battling three used Chelsio T580-LP-CR's that I picked up recently and wanted to see if I'm missing anything. I've installed these cards on my small network, which is set up with a Windows 10 box as my primary workstation with conventional 1GbE connections to a few other servers, but for speed, I wish to have two connected as DAC's with these T580 40GbE NICs.

The Windows box specs are as follows:
  • HP Z8 G4
  • x2 Xeon Silver 4110's
  • 128 GB DDR4
  • 1TB NVMe boot drive
  • Current NIC is the T580*
Though I have an array of oddball Dell T7500 boxes built as you'll see in my signature, I am doing all my NIC tests using the newer TrueNAS boxes that are still in the "build" phase (specifically to replace the T7500s).

Specs of the TrueNAS box I'm working in are as follows. This is the one with the DAC connection to Win 10:
  • TrueNAS Core 13
  • Asus Sabertooth X99
  • E5-2620 v3
  • 4GB Hynix PC4-2400 ECC DDR4 (spare from X10 build, waiting on more)
  • No drives or HBAs at the moment
  • Chelsio T580-LP-CR
I realize 4GB is literally half of what TrueNAS needs for actual operation, but I'm not running any drives at the moment; I configured this box mainly for ease of troubleshooting, for the Chelsio cards aren't presently recognized by TrueNAS Scale 22.02.1.

Both cards are in PCIe 3.0 x8 slots, or in slots capable of x8 speeds. Cables are 5m-long QSFP+ NetApp 112-00178's.

Here's the problem:

I've set the static IP's in both Windows and TrueNAS, and in Windows and the cards won't see each other, nor will they ping each other. This is with the same, basic configuration I've used for ages with prior cards: Windows NIC has a static IP of x.x.2.11 and a gateway of x.x.2.10, and the server NIC is x.x.2.10.

Historically, I have not had to set a gateway on TrueNAS (Core or Scale) or the similar NAS4Free/XigmaNAS boxes to get them each to recognize each other.

The Chelsio cards though, won't connect at all. They won't ping, no packets are sent, and Windows sees it as unplugged. TrueNAS Core 13 sees their links as down in the GUI; ifconfig shows the following. Note that I was trying different subnets here, so please disregard the x.x.3.20 address on one of the ports:

Code:
cxl0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 9216
        description: Chelsio #1
        options=6ec07bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,TSO4,TSO6,LRO,VLAN_HWTSO,LINKSTATE,RXCSUM_IPV6,TXCSUM_IPV6,HWRXTSTMP,NOMAP>
        ether 00:07:43:2e:c0:b0
        inet x.x.2.20 netmask 0xffffff00 broadcast x.x.2.255
        media: Ethernet none
        status: no carrier
        nd6 options=9<PERFORMNUD,IFDISABLED>
cxl1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 9216
        description: Chelsio #2
        options=6ec07bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,TSO4,TSO6,LRO,VLAN_HWTSO,LINKSTATE,RXCSUM_IPV6,TXCSUM_IPV6,HWRXTSTMP,NOMAP>
        ether 00:07:43:2e:c0:b8
        inet x.x.3.20 netmask 0xffffff00 broadcast x.x.3.255
        media: Ethernet autoselect
        status: no carrier
        nd6 options=9<PERFORMNUD,IFDISABLED>


Mind, I've checked and re-checked those QSFP+ cables. ifconfig may be reporting "status: no carrier," but each end is locked as they should.

This is what gets me: I bought the Chelsio cards to replace a set of Mellanox Connect-X3 CX354A 40/56GbE cards (set in 40GbE Ethernet mode, not IB, of course) which were connecting 100% without issue with this same configuration and cables, on the same exact computers, with the same exact settings. And even prior to that, the same HP workstation and servers were all fitted with 10GbE Connect-X2's, with the same IP configurations, without issue.

The only reason I pulled the Connect-X3's is because they were barely cresting ~21Gbps when tested in iperf3 (with -P 16) and I couldn't find any bottleneck other than the possibility of the card itself. Also, for whatever reason, the T580's are not recognized in TrueNAS SCALE either, and the Mellanox cards only recognized in Core 13, and not in any v12 BSD OSes (such as XigmaNAS).

After pulling my hair out with the Mellanox cards, I decided to follow what I'd read here on the forum about TrueNAS having better support and overall tuning for the Chelsios, but now I'm stumped and regretting it!

Any advice would be gladly appreciated. Thank you!

-Kurt
 
Last edited:

sretalla

Powered by Neutrality
Moderator
Joined
Jan 1, 2016
Messages
9,703
I've set the static IP's in both Windows and TrueNAS, and in Windows and the cards won't see each other, nor will they ping each other. This is with the same, basic configuration I've used for ages with prior cards: Windows NIC has a static IP of x.x.2.11 and a gateway of x.x.2.10, and the server NIC is x.x.2.10.
Those numbers look fishy to me...

I don't think you should have the IP of the server and the gateway set to the same address... maybe you wanted to have x.x.2.1 for the gateway?
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
Those numbers look fishy to me...

I don't think you should have the IP of the server and the gateway set to the same address... maybe you wanted to have x.x.2.1 for the gateway?

The gateway is only set on the Windows box, and said gateway is the NAS's NIC.

Maybe this will make more sense:

Win 10:
Assigned static NIC IP: 192.168.2.21
Gateway: 192.168.2.20

NAS:
Assigned static NIC IP: 192.168.2.20

Far as I've read for years, this is the usual way one would set up a static direct connection. If I have it wrong, it has somehow worked on the T7500 boxes for nearly a decade. But maybe that's been a fluke. Let me know.

-Kurt
 

sretalla

Powered by Neutrality
Moderator
Joined
Jan 1, 2016
Messages
9,703
That just clarifies it further... why is the NAS the default gateway?

If you're going to set it up like that, the dedicated link needs its own subnet and no gateway specified on either end.

You shouldn't use the same subnet as your other network.
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
That just clarifies it further... why is the NAS the default gateway?

If you're going to set it up like that, the dedicated link needs its own subnet and no gateway specified on either end.

You shouldn't use the same subnet as your other network.

Fair point. I've had it set up that way for so many years, I never actually thought about it. Bear with me, as I'm self-taught in network matters, and while I've read a lot over the years and applied it, a lot has also oozed out of my memory with the passing of time (I really only spend time configuring stuff like this when I'm involved in upgrades like this).

This said, I've removed the gateway entirely, but left a 255.255.255.0 subnet mask. Doesn't make a difference - the Mellanox cards see each other in this configuration, the Chelsios don't.

-Kurt
 

sretalla

Powered by Neutrality
Moderator
Joined
Jan 1, 2016
Messages
9,703
I can't make sense of what you're describing... I think some kind of diagram is necessary for any further help.

What is connected to what on which subnet and with which address... where is your switching/routing happening and where is your path out to the Internet?
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
This help?

server-array.png


The 1GbE network is maintained only for GUI access.

-Kurt
 

sretalla

Powered by Neutrality
Moderator
Joined
Jan 1, 2016
Messages
9,703
Yes, that makes it very clear and easy to see that you need to read this:

You need to choose separate subnets for each of the back to back links.
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
Yes, that makes it very clear and easy to see that you need to read this:

You need to choose separate subnets for each of the back to back links.

Ok, I think I understand. If you could, let me know if this edit makes sense:

server-array2.jpg


If I've missed the point entirely, I'll go back and read through again and do some more research to get this through my head.

Incidentally, I just realized the way I represented the 1GbE network could be misinterpreted, so I've edited it accordingly.

-Kurt
 

sretalla

Powered by Neutrality
Moderator
Joined
Jan 1, 2016
Messages
9,703
I think what you have on the 1Gb network is a switched environment, so a single subnet is fine for that.

The updated design for the back to back connections seems right now.
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
Just a quick update - solved the access problem by replacing the existing QSFP+ cables; I have a suspicion that they were actually QSFP sold to me as QSFP+. The ends were not marked. Granted, they were going up to 21Gbps, so maybe they were QSFP+, but just not compatible.

Wound up getting these instead out of convenience, which do reliably connect the Chelsio cards, and also Chelsio/Mellanox:
(For the record, if that link ever dies, the cable is an AOC-Q3Q3-005 from 10GTek, spec'ed as 56GbE QSFP+ capable, fiber optic (AOC OM3).

This said, I'm now fighting a different problem, which I've been wrestling with ever since working towards this 40GbE upgrade - ~10GbE speeds from server to Windows. Not a TrueNAS issue either, as I've verified it with a live USB of Ubuntu too. Will start another thread on this.

-Kurt
 
Joined
Dec 29, 2014
Messages
1,135
FWIW, I have had not good experience with DAC cables across different manufacturers. Sometimes it works and sometimes it doesn't. I prefer to use SR optics and fiber patch cords when there are different manufacturers in play.
 

cudak888

Cadet
Joined
Dec 13, 2015
Messages
9
FWIW, I have had not good experience with DAC cables across different manufacturers. Sometimes it works and sometimes it doesn't. I prefer to use SR optics and fiber patch cords when there are different manufacturers in play.

Great, more stuff to spend money on :P

Jokes aside, I'll give this a try - would be silly not to do so.

Incidentally, I'm finding many transceivers that are Cisco, Mellanox, etc. compatible, but barely any that claim Chelsio compatibility.

EDIT: I lied, not enough Google-fu. How about this? https://www.amazon.com/CHELSIO-COMM...737762&sprefix=chelsio+sm40g-sr,aps,88&sr=8-1

FYI, if you want to follow along with the rest of the story, I started a thread at STH, seeing as the issue doesn't appear to be related to TrueNAS itself: https://forums.servethehome.com/ind...iperf-to-windows-on-dac-hardware-issue.36765/

-Kurt
 
Last edited:
Top