CEROWRT-1.0 “OCEAN CITY” RELEASE NOTES

THIS DOCUMENT IS OBSOLETE

See the OCEAN CITY page.

CEROWRT-RC6

These release notes apply to RC6 and may or may not apply to earlier release candidates. RC6 consists of:

  • Linux kernel 3.0.4
  • iptables 1.4.12.1
  • wireless-testing 2011-09-14

as well as too many features to enumerate here.

Note that there are desired features that have not yet landed; e.g. IPsec tunnels, DHCPv6-PD, and completion of DNS support. The DNS vision is that you should be able to plug a named machine into your network and its IPv6 address be automatically published into the global DNS without any manual configuration needed whatsoever. This capability has been made to work in prototype form, but is not yet present in CeroWrt. Preliminary performance numbers for IPSEC were about 20MBPS on the WNDR3700, well beyond current commercial offerings costing much more. If you are so inclined, help with any of these would be greatly appreciated!

eBDP is not currently in Cerowrt; we are considering test builds in RC7 of this AQM algorithm. The current tuning of queue lengths may (almost certainly is) sub-optimal and not yet comparable to factory firmware for bandwidth performance.

  1. If you haven’t installed a router yet, the current draft of the web pages on the router can often be seen at: http://cero2.bufferbloat.net/
  2. Performance testing has just begun and you may very well uncover a bug or a problem with default configuration which need to be fixed.
    Note that this means that naive testing may produce unexpectedly poor bandwidth results: for example, when we were testing CeroWrt directly plugged directly into a gigabit network with 10G back haul, we saw poorer Large, Fat Network (LFN) router performance than the commercial firmware on identical hardware, due to our configuration for use in a more typical home environment. This is hardly surprising, as Linux drivers have been tuned extensively for use at that speed in data centers (to the detriment of latency); when re-configured identically to factory firmware, CeroWrt outperformed the commercial firmware in this example.
  3. Reduced TXQUEUELEN accoss all ethernet and wireless interfaces. Since this is a home router, when operating locally, delay is very low, so the queue can be short without losing TCP performance; when operating remotely, the upstream bandwidth limit means the buffering can again be low. Transmit queues only need to be large when you have both high bandwidth, and high delay paths.

    • Reduced buffering in the ethernet switch
    • Reduced driver buffering across all ethernet and wireless interfaces
      This initial tuning is probably too aggressive, and will be tuned further.
  4. There is ongoing performance testing against various smoketests going on in #262
    We are in the VERY early stages of performance testing and have all sorts of variables - oprofile being enabled, HT40 not being enabled by default, queue lengths, lab setup, and problems with the tests themselves … all left to resolve. Please draw no conclusions from the performance test results thus far, and note that rc6 contains the latest and greatest wireless-testing, not tested on bug #262.

  5. CeroWrt is still not independently buildable until the patch set settles some more and more patches pushed into OpenWrt head.

  6. Our current opinion is that HT40 should only be enabled by default at 5ghz only, and not at 2.4ghz, as it makes the router enough more noise sensitive as to probably cause too many people trouble (though will lab benchmark more poorly unless enabled, of course). Differing opinions are welcome about this default choice.

  7. CeroWrt routes rather than bridges between networks. It has an mdns forwarder to handle the case of a home network using MDNS to locate services. “A little multicast can ruin your whole day on wireless”.

  8. Note that the router is configured with default network numbering to use network 172 addresses to try to stay out of your existing network’s way. This may make renumbering if you have an existing static numbering plan in your house somewhat a challenge. We plan changes in RC7 so that the low addresses are left free for static numbering, as that is the most common configuration people have. If you modify your RC6 configuration for interface se00 (found in /etc/config/network) to enable access to low IP addresses, remember that you must also fix all references to 33 in your bind configuration files (found in root@OpenWrt:/etc/chroot/named/etc/bind/master) so that gw.home.lan will work.

  9. For IPv6, we are most concerned about the following bug: #266

    • dibbler is in the RC6 build: volunteers to play with DHCPv6-PD and test would be great
    • If the WAN interface is allocated a routeable IPv4 address (e.g. directly plugged into your broadband gear) the router will turn on 6to4 by default and advertise IPv6 routes. Please give us feedback as to whether this causes trouble. Comcast has put in many more geographically dispersed 6to4 relays so using 6to4 is much less problematic than it was even a year ago often exhibiting very good performance and reliability on that network.
  10. The eBDP algorithm for wireless queue management is not yet present in CeroWrt builds; it awaits some further testing.

  11. QoS enabled by default; you should tune your QOS settings for your connection as covered in the OCEAN CITY FAQ, or you will have performance problems of some sort.

  12. Something like 20+ new packages are now available, including dibbler, gpsd, nuttcp, pimd, and ccnx.

  13. Mesh routing is in CeroWrt, using the Babel protocol using bandwidth diversity and ahcp for address distribution.

    • Note that the router does not warn you in some trivial way (blinking light, highlight on the status page) that it may be meshed rather than using its WAN port. Feature #270
    • Babeld updated to port 6696 and version 1.2 (incompatable change)
  14. Bugs #195 (ethernet unaligned transfers), #240 (ssh problems), #243 (iptables), #256 (ntpdate) are believed fixed in RC6 and are being tested.

  15. Bug #205 is a PITA for DNS lookup; the current workaround may slow boot by several minutes; there are fixes possible, but not in time for RC6

  16. Babeld moved to IANA port 6696 - this is a non-backward compatible change

  17. Addition of TCP background (BP) and TCP yeah algorithms

  18. Web10g support - note: we have not found a real use for this and may pull it from the next release

  19. Tcp_low_latency made the default

  20. IPv6 works now on the secured and unsecured interfaces by default

  21. Basic support for the cosmic background bufferbloat detector

  22. The roadmap has other known issues scheduled for resolution in RC7; please check that list before starting a new bug report

To edit this page, submit a pull request to the Github repository.
RSS feed

Recent Updates

Oct 20, 2023 Wiki page
What Can I Do About Bufferbloat?
Dec 3, 2022 Wiki page
Codel Wiki
Jun 11, 2022 Wiki page
More about Bufferbloat
Jun 11, 2022 Wiki page
Tests for Bufferbloat
Dec 7, 2021 Wiki page
Getting SQM Running Right

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".