[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: "Niceness" for network transfers
From:       Thiago Macieira <thiago.macieira () kdemail ! net>
Date:       2005-02-17 19:50:59
Message-ID: 200502171750.59991.thiago.macieira () kdemail ! net
[Download RAW message or body]

[Attachment #2 (multipart/signed)]


Rainer Wirtz wrote:
>My question to the geeks: Is there any provision for this functionality
>in IPv4 and/or IPv6? And, if yes, what are the chances for this to
>become available on kde-application level. And what could I do to help
>accelerating this.

IPv6 has something you could call a distant cousin of what you want: 
flowlabels. In theory, they can be used to prioritise traffic. (More or 
less like IPv4's TOS, but much more flexible).

In practice, no router prioritises according to flowlabel/flowid.

The solution is to try and and add limiting filters to your traffic 
control/shaping routines. This is probably the best.

Having said that, however, there's a way we can do that in userspace: by 
allowing the kernel socket buffer to fill up, the kernel will stop 
sending ACKs for received packets, which will in turn make the sender 
throttle down it's output. Therefore, by carefully controlling the read 
rate (kernel -> application), we can more or less control the bandwidth 
used.

I have never tried doing that, but my gut feeling is that it would work. 
Note: you cannot use the buffered socket classes for this, because their 
job is to read from the kernel as fast as they can.

-- 
  Thiago Macieira  -  thiago (AT) macieira (DOT) info
    PGP/GPG: 0x6EF45358; fingerprint:
    E067 918B B660 DBD1 105C  966C 33F5 F005 6EF4 5358

2. Tó cennan his weorc gearu, ymbe se circolwyrde, wearð se cægbord and se 
leohtspeccabord, and þa mýs cómon lator. On þone dæg, he hine reste.

[Attachment #5 (application/pgp-signature)]

>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<


[prev in list] [next in list] [prev in thread] [next in thread]