Oracle ZFS Storage ZS3-2: only X2129A and X5562A-Z are listed for the X1109A-Z 10GbE NIC
We took over a pair of ZS3-2 controllers together with the shelves behind them, and I am the one who has to order optics before they go back into service. The controller service manual is very short on this point: it lists the NIC and HBA options, and for the 10GbE SFP+ card it names exactly two transceivers and stops there.
- Oracle ZFS Storage ZS3-2 controller, 2-port 10GbE SFP+ NIC (X1109A-Z)
- intended use: short-reach 10G uplinks to the top-of-rack switches, multimode patching
- what I have in the rack already: a tray of generic 10GBASE-SR modules used on the switch side
The whole of what the manual gives me:
NIC / HBA options, ZS3-2 controller
X1109A-Z 2-port 10GbE SFP+ NIC
7100477 4-port 1Gb copper UTP NIC
7100488 2-port 10Gb Base-T NIC
SG-SAS6-INT-Z, 7103791 SAS-2 HBA
SG-XPCIE2FC-QF8-Z, 7101674 8Gb / 16Gb FC HBA
X4242A 2-port InfiniBand CX2 HCA
SFP+ transceivers listed for the 10GbE NIC
X2129A 850 nm, 1/10 Gbps, short reach
X5562A-Z 10GbE / 1GbE long reachBefore asking I read the options section end to end looking for any sentence about non-Oracle optics, and there is none in either direction: no permission, no prohibition. I also went through the appliance interface hoping for soomething that prints an accepted-modules list, and did not find one.
So which is it - a qualified-parts list for ordering and service, or a whitelist the card enforces? Has anyone actually put a generic SR module into an X1109A-Z and seen what the appliance makes of it?
Comments 4
Before anyone starts guessing: confirm which card is actually in the slot. The 10GbE SFP+ NIC and the 2-port 10Gb Base-T card (7100488) get mixed up in these chassis constantly, and only one of them takes a module at all. Read the part number off the bracket, not off the manual.
Second thing worth stating is what the port has to do. X2129A is the dual-rate 850 nm part, so it covers 1 Gbps as well as 10 Gbps on short reach, while X5562A-Z is the long-reach one. If the switch side is fixed at 10G SR then only the first of those two lines matters to you, and the question narrows to one module type.
That list is a qualified-parts list and nothing more. It is what ordering and field service work from - the parts an engineer is allowed to ship you, and the parts the appliance was tested with. A page that says nothing at all about third-party optics is neither permission nor a ban, it just does not cover the case.
There are two separate questions hiding in yours. Whether the port lights up is decided by the NIC and its firmware, not by the appliance software on top. Whether you get help when the link misbehaves is decided by support policy, and appliance vendors are consistent there: the first instruction on a flapping link will be to fit a listed part and reproduce.
Compare with a vendor that spells it out. On the Firepower 2100 boxes Cisco lists the optics per port group and then states plainly that non-Cisco SFPs are not blocked but are untested, and that TAC may decline an interoperability case that involves them. Oracle simply never wrote that paragraph for the ZS3-2. Practical answer: keep one X2129A per controller in the spares drawer so you can always fall back to a listed part, and run whatever you like the rest of the time.
Barracuda is the clearest published example of exactly what you describe. Their CloudGen Firewall network modules are documented part by part - ET074 with 8x 1GbE fibre SFP and ET075 with 8x 1GbE RJ45, both on an Intel I350-AM4; ET076 with 4x 10GbE SFP+ and ET091 with 2x 10GbE SFP+, both on an Intel X710; ET085 with 2x 40GbE QSFP+ on an XL710-BM2. And then they say outright that their transceiver testing follows the Intel driver test list rather than an approved-optics list of their own, admit they cannot test everything on the market, and ask users to report the module name, firmware and serial of parts they have working.
Same conclusion as yours: the appliance is not the thing making the decision. A module coded for another host vendor gets refused down at the NIC, and no amount of reading the appliance documentation tells you that in advance.
Worth spelling out where that NIC-level check comes from, since on an Intel-based card it is the whole story. Intel qualifies its own modules per adapter family - E10GSFPSR is the dual-rate 10GBASE-SR / 1000BASE-SX 850 nm part, E10GSFPLR the dual-rate 10GBASE-LR / 1000BASE-LX one - and the adapters read the module EEPROM and by default keep Rx and Tx disabled for anything that is not on the qualified list.
The escape hatch depends on the driver, not on the card. ixgbe has an override:
i40e and ice, which is what X710 / XXV710 / E810 run under, have no equivalent, so there you are back to Intel-coded modules or to patching what you have. I am going off Intel's own compatibility note rather than a card on my bench, so verify the behaviour on your own hardware before you plan a purchase around it. And none of it transfers to the ZS3-2 card until somebody says what silicon is actually on the X1109A-Z.