| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
| Thanks Glen, One of the action item from yesterday's discussion is about actual consumption of ONU. This gave me an idea. I went in our lab and measured the power consumption of a 10G ONU SFP+ 20Km transceiver. I am talking about just the transceiver that can be plugged into an ONU. I could measure the power consumption with Laser ON and with Laser OFF. I repeated the experiment with various brands and got consistent results. Laser | Input Voltage | Input Current | Output Optical Power ------+---------------+---------------+--------------------- OFF | 3.3V | 0.36A | -50dBm ------+---------------+---------------+--------------------- ON | 3.3V | 0.50A | +5dBm ------+---------------+---------------+--------------------- The way I measured power is simple: we have a small test board that pretty much consists of a SFP+ cage. I could read the Voltage and Current from the power supply feeding the test board. The test board itself has no active components so that we can safely neglect its own power consumption. This board is also equipped with few dip-switches controlling some input pins of the SFP+ module. In particular, it can control the SFP_TX_DIS pin which allowed me to turn laser ON and OFF. Now let's consider a very simplistic 10G EPON 'power saving' approach where all we do is slow down the polling activity at night. For our calculations, I will assume the following: - night is between 11PM and 5AM (6 hours) - FEC Enabled - Laser ON : 32TQ - Laser OFF: 32TQ - Sync Time: 16TQ - Cost kWh : 12.55 cents (2021 US average) - ONU is polled at constant frequency with force report set - No traffic is going through the ONU during the night period The simplistic power saving approach is to change the polling period from 1ms (very aggressive setting) to 50ms (very slow setting). Let's now compare the cost of doing nothing (keep fast polling) and slow polling during night time. Of course the data will be about existing 10G (not 25G) and restricted to TX laser savings but it will help us getting some idea about what to expect. With a fast polling, we would have a total of 2.16E7 bursts (6h * 3600s/h * 1000burst/s) per night. Each burst being about 95TQ (based on optical overhead I selected). This means that during the night, the laser will be ON for 32.83s (2.16E7burst/night * 95TQ/burst * 16ns/TQ / 1E9s/ns) Compared to laser being OFF for that time, it corresponds to a cost of 5.29E-5cent/night (12.55cent/kWh * 3/3V * (0.50A - 0.36A) * 32.83s/night / 3600s/h / 1000W/kW) Same calculation with a slow polling, all numbers are basically divided by 50 since we poll every 50ms instead of every ms. So cost of slow polling at night is 1.06E-6cent/night compared to never turning laser ON at night. So, under the above assumptions, and aggregating the cost for 126 million US households: - Having fast polling at night costs $24.3K per year (5.29E-5cent/night * 365night/year * 126E6households / 100cent/$) - Having slow polling at night costs $486 per year (1/50 of previous number) Again, both numbers are in comparison with never turning laser at night. From this, it is not clear to me if implementing this simplistic approach is worth it, nor is it for me to judge but: - It can be implemented easily with existing equipments (Polling period is likely available to network management system). - Its impact on latency is low and controllable. 50ms polling is just an example. - Implementing full blown power saving feature seems to hit diminishing returns unless we can demonstrate that further savings (beyond TX Laser) can be implemented. Please let me know if you see a problem in my experiment or its results. Thanks, Jc
To unsubscribe from the STDS-1904-4-TF list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-1904-4-TF&A=1 |