Since 2024, INX no longer provisions 100gb/s ports using 100gb/s LR4 transceivers. Below is a list of reasons that explains why we do this.
1. Cost
LR1 transceivers are, on average, 25% cheaper than their LR4 equivalents. That might seem marginal on a singlel purchase, but, at scale, the savings are meaningful for a non-profit operating on cost-recovery principles. The delta compounds further when you factor in sparing: we always carry hot spares, and, as Africa's friendliest IXP :-) we also get asked by peers if we can help in a pinch. So, as you can imagine, a 25% reduction in transceiver cost across a sparing pool of any reasonable size adds up quickly.
2. Engineering simplicity and reliability
As you may already know, LR1 is a fundamentally simpler optical design than LR4. Where LR4 uses four discrete wavelengths (1295nm, 1300nm, 1304nm, 1309nm) multiplexed onto a single fibre via a WDM mux/demux inside each transceiver, LR1 uses a single wavelength on a single lane. Fewer moving parts means fewer failure modes.
This is not a theoretical concern; we have already experienced several cases of 100Gb/s LR4 transceivers losing transmission on a single internal lane, which renders the entire transceiver non-functional despite three lanes continuing to operate fine. With LR1, a transceiver either works or it doesn't, which is a far cleaner failure mode to detect, diagnose, and resolve. Additionally, LR1 transceivers tend to have better thermal characteristics and lower power consumption than LR4; which matters at scale.
3. Operational simplicity in the field
We think that this is a largely an under-appreciated point. Data centre technicians, and network engineers still coming into their own, do not routinely carry equipment capable of testing individual lane loss on a multi-lane transceiver. When a circuit connect to a LR4 transceiver fails due to a single degraded lane, the technician's test equipment typically still shows light present, which leads to the inevitable and frustrating conversation: "there's light, so the fibre is fine!" when in reality, the issue is a single failed wavelength inside the transceiver itself.
Explaining PAM4 modulation, per-lane power budgets, and WDM lane mapping to a data centre cross-connect team at 02h00 is not a productive exercise. LR1 eliminates this entirely; light means that there is expectation that the link works, whilst, no light means it doesn't. That binary simplicity saves time, reduces miscommunication, and gets links restored faster. More importantly, it means that the networks team get more sleep! :-)
4. Strategic alignment with our 400Gb/s breakout architecture
In higher-density locations, we provision 100Gb/s ports by gear-wheeling 400Gb/s ports into 4x100Gb/s. This helps us reduces the effective per-port cost significantly (think fewer switches, lower rack space, lower power, etc) and allows us to sweat our 400Gb/s infrastructure more efficiently. This breakout model works precisely because we can optically aggregate four 100Gb/s LR1 transceivers into a single 400Gb/s QSFP-DD port using a 4:1 MUX/DEMUX. This is pretty clean and straightforward with single-wavelength LR1, but would not be possible to manage (easily) with LR4, which already occupies four internal wavelengths per transceiver.
Deploying LR1 at 100Gb/s is therefore not an isolated decision; it is a deliberate, architecturally consistent choice, that underpins our entire 100Gb/s strategy. Introducing LR4 transceivers at this stage would be a regression in thinking, and would create an inconsistent sparing and troubleshooting environment across our deployments. For completeness: when we do encounter LR4 failures in the field nowadays, our replacement policy is to migrate to LR1, and not to replace like-for-like.
5. Future compatibilities
Even if our 400gb/s breakout architecture is replaced by, say, something like SFP112, or if peers start deploying their own SFP112 equipment (we already know networks that are looking at this) we want to be sure we can support this easily. SFP112 and LR1 are optically compatible; LR4, isn't.
So, LR1 provides us with real technical, and architectural value, whilst still being cheaper! It's rare to find technology that does all that, and that's why, in 2024, we decided that were were going to draw a line in the sand, and, to do our technocratic bit to drag, kicking and screaming if we must, operators into the future. Fortunately for most ZA network operators, this actually hasn't been a challenge. We have had pushback from some international CDNs but, we've held our own, and we've gotten every operator that we have engaged with, to move the needle on their own LR1 deployments.
We hope that this captures our thinking, and reasoning adequately. We're aware that other IPXs are more lenient than we are. But, as our INX manager always says, you don't get to run infrastructure with 100% uptime in 30years without being an autocrat of some sort, somewhere. We have several quirks that we don't compromise on (eg. how dogmatic we are about our route filtering policies!); and that we think that these collectively, adds value to INX and her peers!
