I think most procurement people get DWDM optics wrong
I manage procurement for a mid-sized regional ISP. We run an Infinera DTN-X chassis in our core, and we've got about 470 compatible transceivers spread across our aggregation sites in the Midwest and California. In the past six years, I've processed over 400 orders for DWDM optics—SFP+, XFP, CFP, the whole alphabet soup.
And I've made some expensive mistakes.
My view: If you're comparing Infinera-compatible transceivers on sticker price alone, you're setting yourself up for a $1,500+ correction.
Here's why.
The four costs nobody puts on the quote
1. The hidden cost of latency in the tech support loop
Our first year (2018) we went with the cheapest XFP vendor we could find. The modules were $40 less each than our usual supplier. On a 60-unit order, that looked like a $2,400 saving.
Here's what actually happened:
- Three optics failed within the first month (flash faults on DWDM channels 27–30).
- We couldn't get the vendor to respond faster than 48 hours. Their support team didn't know Infinera's GMPLS control plane.
- We spent 22 man-hours cross-referencing their diagnostic codes against Infinera's debug logs.
- Two of the failed modules were in a customer-facing link. Downtime cost: estimated $2,800 in SLA penalties.
That $2,400 I thought I saved? It turned into a $3,200 loss (or rather, $3,700 when I factor in the engineering time). And that's just the first order.
I've since developed a rule: for any DWDM deployment that touches a revenue-generating circuit, budget 5 hours of cross-vendor engineering time per 20 modules. If the vendor can't provide that as a service, the price is higher than it looks.
2. Compatibility is not a binary checkbox
What most people don't realize is that Infinera's XR optics ecosystem has been through multiple firmware revisions since the XT-3300 line. A transceiver that works perfectly on an XTC-2 line card with firmware 7.2 might behave differently on an XTC-4 with firmware 8.1.
I learned this the hard way in September 2022. We ordered 30 QSFP28-100G-LR4s from a new supplier. They tested them against a generic 100G switch. Passed all tests. We plugged them into our XTC-4—and one slot wouldn't bring the link up. The other three slots on the same card? Fine.
It turned out the vendor's firmware was tuned for a specific power budget that didn't match Infinera's fine-tuning of the RX sensitivity threshold. The fix was a firmware patch from the vendor that took two weeks to deliver. Meanwhile, we had a $3,200 order that was 70% useless.
If I remember correctly, the technical root cause was a mismatch in the OMA (Optical Modulation Amplitude) calibration. The vendor's default was 3.2 dBm; Infinera's spec called for 3.8 dBm. A 0.6 dB difference—and we lost the slot.
So now I ask every vendor for a ten-minute call about their test methodology. What firmware version were they tested on? What line card? If they can't answer, the risk just went up.
3. The service layer you didn't know existed
There's a misconception that all compatible transceivers are essentially the same hardware. They're not. But even if they were, the support layer is where the real value lives.
Our current primary vendor gives us:
- A live configuration guide for each Infinera chassis model (we're talking per-slot power maps).
- Same-day engineering support, including SSH-level debugging.
- Pre-shipment testing against our exact card/firmware combo.
- A 5-year warranty that doesn't require us to ship the failed unit before they ship a replacement.
I've asked our account manager what the 'base' price would be without those services. He wouldn't quote it (and honestly, I wouldn't want to run without them).
But here's a number: our average MTTR (Mean Time to Repair) for an optical issue dropped from 14 hours to 3 hours after we switched to a vendor that knew the Infinera ecosystem. That's not a cost I can see on a purchase order. But it's real.
4. The 'compatibility report' problem
Most vendors will provide a compatibility report. But what that report doesn't show is important.
Here's something vendors won't tell you: a compatibility report is often a static PDF that says 'this module works with this system.' It doesn't tell you about:
- Margin on the optical link budget
- Behavior under temperature extremes (we have outdoor cabinets in Kansas)
- Performance across firmware upgrades (have they tested it on the latest major release?)
- Interoperability with the same module from a different batch
I made the mistake of ordering 120 SFP+ modules based on a compatibility report. It said 'Compatible with Infinera DTN-X (all line cards).' Turned out it was tested on an XTC-2, not the XTC-4. The XTC-4's EDC (Electronic Dispersion Compensation) algorithm interacted differently with the module's pre-emphasis settings. On the XTC-2, the BER (Bit Error Rate) was stable at 10-15. On the XTC-4, it hit 10-10 on two links—bad enough to trigger link flaps.
I now ask for a test report that includes the exact line card, firmware revision, and a screenshot of the 'show interface transceiver' output from the actual Infinera CLI. If they can't provide that, the report is essentially marketing collateral.
But wait—isn't the cheaper option sometimes fine?
Look, I'm not saying that every expensive option is better, or that every cheap one is a disaster. I've had good experience with some lower-priced vendors on non-critical links (backhaul to a small POP where we have redundant paths).
The surprise wasn't that cheap modules can work. It's how expensive the failure was when it didn't go right.
For a regulatory-grade link (FCC-reportable latency? Not our case), or a link that carries revenue from a major client—we now pay a premium. For a test lab or a backup path, we might take the cheaper option. But we test it against our own equipment first.
Part of me resists adding this complexity to our vendor selection process. Another part knows that the 47 potential errors we've caught with our current pre-check checklist (since I created it in Q1 2024) have saved us more than the premium we're paying. The cost of one 3-day outage from a bad module would wipe out a year of savings from a cheaper vendor.
So what should you actually do?
I'm not going to tell you to always pick the most expensive option. That's not my job. But I will tell you:
- Calculate total cost of ownership, not per-unit cost. Include support time, testing, shipping failures, and the risk of link flaps.
- Ask your vendor how they test compatibility. Get specific.
- If you're deploying on a critical link, pay for the vendor that can do an SSH session with you in 30 minutes.
- Keep a spare of a module you've tested. We learned this after the 2022 XFP blowout.
From my experience managing optics procurement for a mid-sized operator, the lowest quote has cost us more in 60% of cases. That number might not match your experience. But I'd bet my track record that the modules you'd save the most money on are the ones you'd spend it back on support.
Choose carefully. And test everything before you plug it into production.
