I have installed a new copy of OpenWrt and Luci today. I have BT Fibre 100 (fibre to the house). When I install SQM is slowing down the connection. I have tried both cake/piece_of_cake and fd-code/simple.qos and they both have the same effect.
A typical set of readings are:
without sqm ping 28 download 144.58 upload 29.2
with sqm ping 33 download 136.60 upload 28.63
Thanks for your responses.I have looked at this again.
Before applying SQM my speed test over three tests gave:
ping 25ms Download 143.05Mbps upload 29.00Mbps
ping 29ms Download 141.49Mbps upload 29.16Mbps
ping 32ms Download 146.26Mbps upload 29.13Mpps
Mean values: ping 28.67ms Download 143.6Mbps Upload 29.09
With Queus discpline: cake; Queue setup script: piece_of_cake over three tests
Ping 42ms Download 115.95Mbps upload 24.97Mbps
Ping 33ms Download 111.55Mbps upload 25.69Mbps
Ping 50ms Download 117.52Mbps upload 25.71Mbps
Mean values Ping 41.67ms (+45%) Download 115Mbps (-19.9%) upload 25.46Mbps (-12.48%)
So everything seems worse.
With Queus discpline: fq_codel; Queue setup script: simple over three tests
Ping 48ms Download 115.64Mbps upload 27.3Mbps
Ping 29ms Download 116.77Mbps upload 27.18Mbps
Ping 51ms Download 111.62Mbps upload 27.33Mbps
Mean values Ping 42.67ms (+48.8%) Download 114.68Mbps (-20.1%) upload 27.27Mbps (-6.26%)
So again everything worse. Cake slightly better with upload.
The finally to check nothing else has changed a three tests with SQM off again:
Ping 25ms Download 148.55Mbps upload 29.98Mbps
Ping 56ms Download 145.3Mbps upload 29.18Mbps
Ping 35ms Download 148.92Mbps upload 29.18Mbps
Mean values Ping 38.67ms Download 147.59Mbps upload 29.44Mbps
Hi Eduperez,
Thanks. The only parameters I set were the dowload speed of 130Mbps and upload 27Mbps and on the link layer adaption tab per packet overhead 34. Are these OK?
Thanks,
Chris
Well that depends, but to quickly run the numbers, out of your set gross-shaper rates and the per packet overhead you can expect at best the following goodput (TCP/IPv4, MTU 1500):
real world measurements tend to only approach but not to reach or exceed these values.
So your measured upload numbers for SQM seem a tad high* and the download numbers seem a tad low, the former might might be attributable to rounding and which time window a test actually evaluates (many speedtests, try to ignore the initial ramp-up phase of a connection, and that in turn with a bit of jitter can skew the measured rates). How did you perform these tests, by the way?
The strange part is the higher ping with SQM .
SQM should slightly decrease speed but it should lower ping during load (compared to full load without SQM).
You seem to experience higher ping with SQM, which makes me to think that you have something wrong.
As I understand it, when the ping tests are done without any load on the connection (as most synthetic tests do), then SQM can only make things (a little bit) worse.
Without SQM, I would expect the ping times to get (much) worse under load; with SQM, ping times should not be affected by the load.
This seems right in principle, so we need to figure out how these numbers were generated to try to understand what is happening here.
It would be good to see the output of: cat /etc/config/sqm tc -s qdisc
the result of a speedtest tc -s qdisc
The idea here is to get a glimpse of cake's internal counters before and after a load to see whether cake notices the increased latency at all. And then we can take it from there...
Okay, with these numbers the theoretical maxima are: 130 * ((1500 - 8 - 20 -20) / (1500 + 34)) = 123.05 Mbps 27 * ((1500 - 8 - 20 -20) / (1500 + 34)) = 25.55 Mbps
But that as expected does not change anything really important.
and these
indicate that cake has not seen noticeable internal delays. This is why now a speedtest result and the output of tc -s qdisc after that speedtest would be interesting to look at.
Thanks for your help. Please can I confirm that you want these outputs before and after a speedtest? I am using the BT Wholesale speedtest.
Thanks again,
Chris.
Yes, I want to compare what happens to cake's counters during the test, so the best is to compare the before and after outputs.
I would handle that with extreme care, I just ran a test from a well connected (gigabit ethernet) place, but " DOWNLOAD 1631.36 Mbps" or 1.6 Gbps seems a bit optimistic.
In short please use a better test, like fast.com (do an initial dummy test then click "settings" and changeParallel connections to Min: 16 / Max: 16; Test duration (seconds) to Min: 30 / Max: 30; and check the three checkboxes).