[Bloat] Three new diffserv codepoints suggested

[Bloat] Three new diffserv codepoints suggested

[Dave Taht]

Before I go to the trouble of writing up an RFC, I thought perhaps
here (and the end-to-end list) would be a good place to discuss some
ideas  towards adding 3 new codepoints to diffserv,
in an enhancement to rfc4594[1] - http://tools.ietf.org/html/rfc4594

I've named them with something of a sense of humor, in a ha-ha only
serious manner.

They are 'Information, Text', 'Large Bandwidth', and 'mice'[2].

1) The 'IT' codepoint would be suitable for interactive ssh and
textual chat (irc,jabber,etc), and merely codify existing practice
that uses the 'immediate' TOS field for this purpose in ssh. The OAM
codepoint seems unsuitable for this.

2) The 'LB' codepoint = 0x63 is an indicator that as soon as a packet
marked in this way arrives at a downstream site, it should be
immediately reclassified and shaped to fit the available bandwidth of
the downstream site.

(a suitable nickname for 'LB' would also be 'Lying Bastard', as there
exist many sites that set all the diffserv bits in the hope of getting
better service from elsewhere)

3) The 'MICE' codepoint consists of small, infrequent, but system
critical packets such as ARP, DHCP, various forms of ICMP needed for
ipv6 ND and RA to work, and DNS.

Arguably the MICE codepoint could be CS6 - the RFC recomends that NTP
be an EF service and says that (in section 4.9 of the standard) that
DNS, DHCP, and bootp must be 'best effort' services classes.

I will certainly argue that wireless is showing that the latter 3
packet types - and ARP, ND, RA, etc -
deserve a class of their own. CS6 could be used to wedge most of these
together, except for the elephant in its room, BGP, and the conflation
with longer-haul protocols. NTP consistently treated as EF would be
nice, too.

thoughts? What other unofficial codepoints are in use? Does anyone
still care about diffserv marking? is the core all MPLS? The edges are
certainly not using anything at the moment.

-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://the-edge.blogspot.com

1: it was nice to run across at least two members of this list that
authored this RFC, and the idea of a MICE class comes from kathie
nichols pounding into my head that the mice are a problem.

2: I was unable to come up with a back-acronym for 'mice'. I got as
far as 'majorly important' before getting stuck on the 'c'.

Re: [Bloat] Three new diffserv codepoints suggested

[Dave Taht]

So I sat down and took the fragments of various qos scripts I'd been
fiddling with, and came up with a diffserv
classifier (doesn't do qos or cope with nat yet) that matches a great
deal of traffic somewhat sanely (notably 'IT', 'mice', and ipv6). It
should work on any linux box, as well as openwrt - (but note nat & qos
caveat above)

It's in my github repo at:

https://github.com/dtaht/Diffserv

(send me your login, I'll give commit access, or you can just do a:

git clone git://github.com/dtaht/Diffserv.git

cd Diffserv/scripts

sudo ./diffserv_dbg up # to turn it on
sudo ./diffserv_dbg status # to see some statistics

Various iptables rules for ecn matches, etc, in there, too.
(the ecn match for ipv6 requires netfilter-head)

While the huge problems of classifying traffic on a large scale on the
edge seem almost tractable with crowdsourcing, still the plethora of
random sources on port 80 seem impossible to cope with.

One thing I intend to use this for is to be able to track delays
through the network. I can now filter on a given
traffic class during real traffic, during a tcp dump, with and without
qos enabled, with or without debloating techniques in place.

-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://the-edge.blogspot.com

Re: [Bloat] Three new diffserv codepoints suggested

[Jonathan Morton]

I thought: Most Important Control Engrams. 

Also, Linux 3.0 appears to be getting QFQ which might be a better alternative to SFQ. I've read the paper and it seems to be both clever and sufficiently simple to work. 

The key to knowledge is not to rely on others to teach you it. 

On 11 Jun 2011, at 16:15, Dave Taht <dave.taht@gmail.com> wrote:

> I was unable to come up with a back-acronym for 'mice'. I got as
> far as 'majorly important' before getting stuck on the 'c'.

Re: [Bloat] Three new diffserv codepoints suggested

[Dave Taht]

On Mon, Jun 13, 2011 at 8:10 AM, Jonathan Morton <chromatix99@gmail.com> wrote:
> I thought: Most Important Control Engrams.

Works for me. I have the classifier up at my git repo, I hope to figure out
how to get this sort of stuff into 802.11e and 802.1d next.

>

I came up with a mildly more apt name for 'Information, Text', calling
it 'Immediate Text'. There's an even better name for it:

'Bandwidth Only For Help'

For which I doubt the acronym for will get past the first PHB, but *I*
like it...

Also, I gave up, and created a new PTP diffserv class for torrents, etc
using codepoint 9 which basically combines CS1 (bulk) with the TOS bit
of MMC (minimize monetary cost)

so, four new codepoints that adaquately classify a great deal of traffic
that has come to light since the original RFC...

> Also, Linux 3.0 appears to be getting QFQ which might be a better alternative to SFQ. I've read the paper and it seems to be both clever and sufficiently simple to work.
>

I'll read it. I only just this morning get DRR into the cerowrt build...

> The key to knowledge is not to rely on others to teach you it.

Well, right now I'm trying to move Linux forward from 1998...

>
> On 11 Jun 2011, at 16:15, Dave Taht <dave.taht@gmail.com> wrote:
>
>> I was unable to come up with a back-acronym for 'mice'. I got as
>> far as 'majorly important' before getting stuck on the 'c'.
>



-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://the-edge.blogspot.com