Updates using idle bandwidth

Les Mikesell lesmikesell at gmail.com
Wed Mar 26 12:48:22 UTC 2008


Tomasz Torcz wrote:
>
>>  > Module tcp_lp.ko will be autoloaded by kernel. So this setsockopt() call
>>  > is ready to be put into yum-updatesd, just after socket creation.
>>
>> A good question then is how do you set congestion level off and not get the 
>> kernel module loaded (if its not needed you wouldn't want it loading just to 
>> have it do nothing).  Does setting TCP_CONGESTION = 0 result in not loading the 
>> module or does it just not limit traffic?
> 
>   That's not how it works. TCP has mechanisms for dealing with congested
> links. There is *always* some algorithm present in TCP stack which
> manipulates connection parameters to not overflow links. By default its
> reno (or cubic), but Linux kernel provides several others to *choose*
> from. This isn't a question of turning limiting ON or OFF.
> 
>  Anyway, after some more reading I see that TCP Low Priority is
> sender-side algorithm. I don not know if sending ACKs is impacted by
> congestion algorithms, and this is only thing for client to modify.

Yes, yum is mostly receiving.  An application can control receive 
bandwidth simply by rate-limiting its reads from the socket, but it will 
have no idea what an appropriate rate might be.

What bandwidth constrained sites with more than one user really need are 
caching proxies, but the fedora mirrorlist configuration will end up 
picking different URLs to get the same file each time instead of using 
copies already in the cache.

--
   Les Mikesell
    lesmikesell at gmail.com





More information about the fedora-devel-list mailing list