Bug #335

Performance regression with rc6 vs bql-34

Added by Dave Täht on Feb 7, 2012. Updated on Apr 21, 2012.
Closed Normal Dave Täht

Description

I had at one point (or so I recall) been able to coax 400+Mbit out of the router in packet forwarding
performance. Exactly how I did that, it was back in august, and I was using higher
end servers and cards then than I am now.

With bql-34 and my current setup I get only 124Mbit in forwarding performance between two laptops
with bql on them. This is with kernel 3.2.5 with bql backported.

It isn’t the new qdiscs.

I regressed to rc6 and tried again, only to get 183Mbit that way, using iperf.
rc6 was 3.0.4 based.

Admittedly I spend most of my time trying to get good 4 mbit performance, but I do
find this puzzling and this strongly indicates a need for a new oprofile.

I am also crashing the router with heavy wireless traffic.

History

Updated by Dave Täht on Feb 8, 2012.
BQL-36 gets 240Mbit over ipv6 with the default ip6tables rules enabled.
280 with all the ip6tables rules disabled.

204Mbit with bql set to 9000 and tx 64 and sfqred in place
234Mbit (no typo!) with tx 2

tc -s qdisc show dev ge00
qdisc sfq a: root refcnt 2 limit 300p quantum 1514b depth 127 headdrop divisor 16384 perturb 60000sec
ewma 3 min 4500b max 18000b probability 0.2 ecn
prob_mark 141 prob_mark_head 7363 prob_drop 20
forced_mark 0 forced_mark_head 0 forced_drop 1194
Sent 1863591533 bytes 1247669 pkt (dropped 1214, overlimits 8718 requeues 134180)
rate 624bit 1pps backlog 0b 0p requeues 134180

d@ida:~/t\$ iperf -t 60 -w256k -V -c fd43:e2b3:341c:f010::2
————————————————————
Client connecting to fd43:e2b3:341c:f010::2, TCP port 5001
TCP window size: 256 KByte (WARNING: requested 256 KByte)
————————————————————
[ 3] local fd43:e2b3:341c:1:214:d1ff:fe20:10cb port 38806 connected with fd43:e2b3:341c:f010::2 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-60.0 sec 1.64 GBytes 234 Mbits/sec

Updated by Dave Täht on Feb 8, 2012.
OK, found it. depth 10, in this case, will cut performance to 170Mbit.

So I think during that testing I’d been fooling with the depth.

Client connecting to fd43:e2b3:341c:f010::2, TCP port 5001
TCP window size: 256 KByte (WARNING: requested 256 KByte)
————————————————————
[ 3] local fd43:e2b3:341c:1:214:d1ff:fe20:10cb port 39253 connected with fd43:e2b3:341c:f010::2 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-60.0 sec 1.20 GBytes 171 Mbits/sec

Updated by Dave Täht on Apr 19, 2012.
massive improvements by #360
Updated by Dave Täht on Apr 21, 2012.

This is a static export of the original bufferbloat.net issue database. As such, no further commenting is possible; the information is solely here for archival purposes.
RSS feed

Recent Updates

Jul 21, 2024 Wiki page
cake-autorate
Jul 21, 2024 Wiki page
What Can I Do About Bufferbloat?
Jul 21, 2024 Wiki page
Tests for Bufferbloat
Jul 1, 2024 Wiki page
RRUL Chart Explanation
Dec 3, 2022 Wiki page
Codel Wiki

Find us elsewhere

Bufferbloat Mailing Lists
#bufferbloat on Twitter
Google+ group
Archived Bufferbloat pages from the Wayback Machine

Sponsors

Comcast Research Innovation Fund
Nlnet Foundation
Shuttleworth Foundation
GoFundMe

Bufferbloat Related Projects

OpenWrt Project
Congestion Control Blog
Flent Network Test Suite
Sqm-Scripts
The Cake shaper
AQMs in BSD
IETF AQM WG
CeroWrt (where it all started)

Network Performance Related Resources


Jim Gettys' Blog - The chairman of the Fjord
Toke's Blog - Karlstad University's work on bloat
Voip Users Conference - Weekly Videoconference mostly about voip
Candelatech - A wifi testing company that "gets it".