SG-2440 Enabling shaping cuts download speed in half
-
I had mentioned this in an post earlier this year but this same client would like shaping enabled now.
Cable ISP is officially 300 down, 30 up. I can test at about 340-350 down on most days.
If I run the traffic shaper wizard, use PRIQ, and prioritize a SIP alias, prioritize DNS, and make POP/SMTP lower priority, speed tests run at around 130 down, 11 up. If I click the Remove Shaper button and run the speed test again I get 245 down, 32 up so obviously faster. I've gone back and forth several times back then, and now; it's not random other traffic.
CPU is not maxing out, and I have disabled Suricata. I also tried unchecking "Disable hardware checksum offload" which is recommended for Suricata.
I also tried download a 5 GB file from Microsoft so not just speed tests.
Suggestions?
-
I banged on this a bit more today and saw less of an effect from shaping, which seems odd that it should change. I do think their neighborhood is seeing high traffic on the cable system as speeds are inconsistent, something I've seen before. Notably today was Saturday at 4pm vs Friday at 5:30 closer to the start of peak evening.
In a few months we are likely getting them a 6100 and moving the 2440 to a branch office so maybe that will improve things. I dislike throwing hardware at a problem but they will need a router in the other location anyway.
-
Tonight after a handful of tests at their location, at 5pm, I notably found Google's speed test about half the speed of others. So definitely the speed test used matters.
-
@steveits
That's the problem with doing speed tests. They can artifically be good or bad. Real world testing, in my opinion, is the only truly valid test. Of course this give a bit of subjectivity "It feels faster".Traffic shaping on asymmetric links is always difficult. I think that traffic shaping is outbound, so make sure all the numbers reflect the uplink values. 30 up, there is a bit of overhead, cable modem is maybe 10% or so (I think, going by memory), so use say 25M for the bandwidth.
I've not setup traffic shaping on pfSense, so all the above is based on different documentation on pf itself.
-
@mer I'm using PRIQ which, per the Netgate hangout video, prompts for speeds then doesn't use them. I agree speed tests need to be stable to be relevant. This is one of the most unstable I've seen in terms of max speed.* But on several different days now, months apart, I went back and forth several times in a row turning shaping on and off and it definitely seemed well correlated. And my alt test is downloading a 5 GB file from Microsoft or someone...after 15-30 seconds one can generally get an idea of that speed. (Exchange CUs are that size)
Ideally, also don't test from 5-11 pm which is prime Netflix time.
Thanks for answering though...I've posted a few issues in this subforum in the past couple years and rarely get a response. :)
*This is a church with a business connection, but in a mostly residential neighborhood near downtown. After they upgraded speeds this spring I spent several hours there one afternoon troubleshooting a "slow connection." The speed tests were slow but ultimately the issue was testing from this person's old Mac (wired) was dramatically slower than me plugging my laptop into her patch cable, or testing from any other computer. No idea what's going on there other than a problem on the Mac but it was frustrating that the speed testing back then was unstable. (of course the client complains about the Mac too but doesn't want to replace it)
-
@steveits
Yep, a lot of people forget that "broadband" is a shared pipe (at least for cable) and "speeds" are not guaranteed. "We'll provision you for that but you may not always get that".I think the traffic shaping is tied to ALTQ stuff in the kernel, if you look at the flow of data from an application, down through the stack to a device driver thence to the wire.
Some ethernet devices have the ability to pull packets from different queues (DMA rings) to put on the wire, so the best traffic shaping is going to be supported down to that level. It all depends on how the queues on the stack work and are processed.That's a kind of long winded way of saying "hardware may help, but it may not"
-
@mer I was convinced it was the shaping on Friday evening. I am slightly less so now, but I am not inclined to stay up to 4am to do more speed testing. :) Of course I'm not on location to plug into the Comcast modem but I did when I was there. We have not had any issues with PRIQ shaping and speed at any client before.
I did with CBQ and found the wizard doesn't set the borrowing to share the parent connection, which has some speed limiting affects on the child queues. Ultimately we moved everyone to PRIQ. And there are other issues with the wizard which I suspect are bugs like the priority of qOthersLow is higher than the default queue.
-
I'm seeing the exact same thing. Every time I enable the Shaper my speeds decrease rather than increase.
This post talks about adjust the qlink, but isn't the point of the Wizard to take this out of our hands? https://forum.netgate.com/topic/162740/qos-traffic-shaping-issue
-
@ksignorini That is me replying there, and is this same router. :) I am not very sure the wizards are that great...like I mention there it seems to use a mix of units, and elsewhere I've posted that the "low" queue seems to be a higher priority than the default queue. Back then I got 90-100 Mbit and currently get 130-260 depending on the test site.
"borrow" is for CBQ and I think the wizard defaults are incorrect there also as it doesn't have borrowing set on the parent so I had a lot of trouble getting expected speeds on CBQ. Eventually just started over and changed everyone to PRIQ and have been fine.
-
Thanks. I'm quite new to pfSense, so I'll continue to monkey with.