[LARTC] Understanding bandwidth in HTB

[LARTC] Understanding bandwidth in HTB

[Leszek A. Szczepanowski]

I have a HDSL at 2048 kbit. The maximum transfer I can
achieve is about 237 kB/sek (using for example FTP or
NetBIOS transfers in my neighbourhood). That is 1941504 bits
per second, 1941 kbit. Where about 106 kbit disappeared ?
Is this protocol overhead ? HDSL framing+ethernet+IP+TCP ?
If so, what values I have to use at designing good HTB tree?
May I suppose that my link is "only" 1941 kbit ? Now there
is 2048 kbit in my script and at high loads HDSL seems to
be overloaded... Even high priority class for ICMP doesn't
work well. Buffers overrun? Is HTB consider protocol overheads,
or we have to consider it individually?
-- 
    Leszek A.      [The chance of a lifetime is passing you by.]
  Szczepanowski    [ Use it or lose it in the wink of an eye.  ]
twinsen@plusnet.pl [ Turn around, giving more than you take.   ]
    gg:275696      [       Lost and found in the past.         ]

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Understanding bandwidth in HTB

[Stef Coene]

On Thursday 27 February 2003 22:51, Leszek A. Szczepanowski wrote:
> I have a HDSL at 2048 kbit. The maximum transfer I can
> achieve is about 237 kB/sek (using for example FTP or
> NetBIOS transfers in my neighbourhood). That is 1941504 bits
> per second, 1941 kbit. Where about 106 kbit disappeared ?
> Is this protocol overhead ? HDSL framing+ethernet+IP+TCP ?
Yes it can be.  You can try to read the counters from iptables or your network 
card or tc to get an idea about the real speed you are sending/receiving.  On 
www.docum.org on the monitor page, you can find some scripts that can monitor 
iptables rules or tc counters.  So you can see what you really are sending.

> If so, what values I have to use at designing good HTB tree?
> May I suppose that my link is "only" 1941 kbit ? Now there
> is 2048 kbit in my script and at high loads HDSL seems to
> be overloaded... Even high priority class for ICMP doesn't
> work well. Buffers overrun? 
If you are sending more then your HDSL modem can handle, it will queue all the 
packets in a hugh buffer.  This can be used to speed up downloads, but it's 
bad for latency.  So if you never send more data then your modem can handle, 
these queues will allmost be empty.  However, it's not easy and takes a lot 
of trial and error to find out how much data your modem can handle.  If the 
modem has small queues, you don't have to worry about sending too much data.

> Is HTB consider protocol overheads,
> or we have to consider it individually?
HTB works for each 0 and 1 that you send.  So including packet headers, 
protocol overhead, ...

Stef

-- 

stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.oftc.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/