Successful 400 Mbps fq_codel SQM with an old Celeron G530

No need for help or anything, just wanted to record my experience somewhere public in case it is useful for anyone with similar hardware:

  • I set up OpenWrt using an old Sandy Bridge-era Celeron G530 (2C/2T, 2.4 GHz, 65W TDP)

  • Despite its age, the Celeron has enough power to do fq_codel SQM at 410 Mbps down / 21 Mbps up. I used simple.qos.

  • The line speed from my ISP is around 475 Mbps down / 25 Mbps up. After a lot of tweaking, I found that setting the SQM speed beyond ~410 Mbps down / 21 Mbps up introduced the chance of latency spikes. While I could still maintain an A+ bufferbloat rating from DSLReports up to around 440 Mbps, my own local ping testing did show some latency spikes at these speeds, so the 410/21 setting is the highest level where I did not see any spikes through my own teting.

  • I initially tried cake with piece_of_cake.qos, but found that regardless of the speed I set, there would always be a few seconds where latency would spike when the download or upload test began, almost as if it took a few seconds before the SQM realized it needed to take effect. Switching to fq_codel/simple.qos made this initial spike go away.

  • The Celeron did not struggle even when applying SQM at 450 Mbps, which indicates that faster transfer speeds are possible with this CPU, if your ISP supports it.

That's all... hope this is useful for someone out there!

1 Like