[gentoo-user] Re: traffic shaping

[James]

Sven Köhler <skoehler <at> upb.de> writes:


> i would like to shape the traffic of my DSL-connection, but somehow i
> never really understood the machanisms that linux offers. All the
> scripts i wrote were simply worthless somehow, because they didn't
> really improve anything.

> Is there any application or script that is easy to configure and does
> all the necessary things to shape my DSL traffic?

First you need really good network management and monitoring. Nagios,
big brother, and Jffnms are just a few of the choices (jffnms is my choice)
and an ebuild is available. Network monitoring will ensure you do not have 
any bottlenecks behind your ADSL connection. The first/best thing you 
can do is to monitor your internal traffic, to figure out where you need 
to concentrate (hardware and protocols).

Then you optimize your firewall/bridge/router to what works best for you.
Last you look external, to optimize what you are getting and see if what 
you are doing can be 'outsourced' like DNS services. Having multiple 
secondary dns servers, strategically located around the internet can 
be a big help, particually if you are serving up lots of small datagrams 
to a large variety of web-surfers. 


You may also be able to pay the carrier/provider some
more money and get more bandwidth. Inquire about CIR or Commited 
Information Rate (guaranteed bandwidth) as some aDSL providers will 
increase  this setting, for a few extra dollars per month. Many 
aDSL providers have greatly over subscribed
their connection from their interexchange points (where they hand off 
internet traffic to other networks). These problems are difficult to 
diagnose, and may be transient. Other aDSL providers such as large 
brain-dead carriers set the CIR to Zero, thus giving everyone great
bandwidth(reletively) but then during peak demand their networks clog will
collisions. Unfortunately, particulary here in the US, the carriers 
are quite stupid and have 'pink slipped' most of their computer scientists 
and  electrical engineers, and they have hired more sales,
marketing and data-base weenies.... Lots of folks that do not know
anything about communications.

aDSL suffers form another unique hardware problem. The current 
( available power) is limited if you are in a wire bundle that is 
carrying lots of aDSL or digital information. Often, when somebody orders a
service, such as aDSL or ISDN, the tech sorts thru the cable pairs at the
terminal blocks down the street and finds the cleanest pairs for the 
new circuit. They migrate the olders services to 'wiring pairs' of lesser
quality. Often the problem is due to corrosion on the terminal blocks. The
bottom line is these carriers are full of idiots and persons not 
qualified to run a network. Here in the US, we're down
to 3 carriers, and they all SUCK. If these idiots can't maintain their
wiring infrastructure, will you trust them to run swithes, routers
and DACs? Many of these carriers are plagued by swams of critical services
that run on the MicroSuck operating system.....

I finally tracked down a problem with my cable modem bandwidth to the RG-59
cable from my home to the terminal block. The dB reading were horribly low.
I had to supply RG-6 (better grade of coax) cable to the technician 
because the cable company is too cheap, stupid, and has no competition. 
If you are nice  to the technicians some of them will help you 
'off the record'. If helps if you can develop a business/personal 
relationship with the tech and the persons that run the equipment.
(I digress much... but this hits  a nerve with me....) There 
is absolutely nothing wrong with aDSL, except the idiot
managers that runs these global communications companies and their 
consistent string of bad decisions.

Monitoring the Internet, is not a bad idea, if you have the time. I can't
remember the name of the software, but it is burried somewhere in the 
NANOG archives.... If what you find is outmoded (based on something like
traceroute) then just find a good security hacking site, as those guys 
stay up on the latest in global network monitoring....

Outside bandwidth testing is possible via these sites:
http://www.speakeasy.net/speedtest/
http://web.tampabay.rr.com/giis/
http://www.dslreports.com/stest

Many others exist; take these results, with caution as to the
accuracy and validity.

This is a deep subject, that is affected by many things. If you build your
own router on a 2.6 linux kernel, there are many things to consider and 
test. 

Beside how you implement your firewall/bride/router rules, there are 
other things you can adjust, under the Networking:QoS and fair queuing, 
when you build thekernel for your primary device connected to the aDSL
connection. PPPoE can be problematic (i.e. not very efficient) if the 
aDSL carrier has a poor implementation or overloaded the resources on 
their side.

The simple solution, is to 'pony up more cash per month' to your aDSL
carrier, or find another bandwidth supplier (if that an option).
You might also have a wireless internet operator in your area. 
Depending on your network you may be able to migrate/split your traffic 
across multiple connections.... 

One of the other respondants mentioned HTB, Hardware Token Buckets, 
which is really a very cool, but a very young technology in Linux. I'm 
still looking for  accurate bandwidth mesuring/monitoring software, 
based on HTBs:

http://luxik.cdi.cz/~devik/qos/htb/manual/theory.htm

I refer to them as Hardware Token Buckets, because that's where software,
meets hardware, in the general area of firmware. The linux scheduler 
choices,even RTlinux hacks are actually piss-poor, when you compare 
their performance
to that of state machines or an efficient RTOS.   The decision of how
a schedulers is implemented in a state machine/RTOS/Kernel effects the
performance of what you really want to acheive. Intimate knowledge of the
underlying hardware (processor) is the key to efficient implementation of
schedulers and HTBs. Many of the developers working on providing HTBs in 
Linux, really need help from real hardware engineers..... But 
eventually, the  implementation of HTBs will be mastered by the 
linux kernel folks......

Obviously a stong, jaded opinion, apologies in advance to all I have
offended, with these truths.....



HTH,

James



-- 
gentoo-user@g.o mailing list