up2date, mirror repositories, and performance

[Jim Cornette]

Martin Alderson wrote:
> 
> On 18 Feb 2004, at 22:37, seth vidal wrote:
> 
>>
>>> Ok, it would mean running bittorrent alongside thttp. Not a huge drain
>>> of resources.
>>
>>
>> What are you talking about? I run a mirror AND I run the torrent tracker
>> for fedora releases. Guess what - the tracker and the seed eat an
>> enormous amount of the total cpu time on that machine.
>>
>> Much higher than just apache or thttpd.
> 
> 
> No, there would be a dedicated tracker. There is no reason why they 
> couldn't be dedicated seeds as well (as opposed to leapfrogging of 
> current mirror). It would just depend on the scale and size of the 
> BitTorrent network apart from FTP and HTTP updates.
> 
>>
>>> Sorry but if it means 20MB that the mirrors don't have to provide (IE:
>>> from users upstream connections) it is a huge success IMO.
>>
>>
>> The mirrors are still going to take a primary hit as the seeds - and
>> fast seeds at that.
>>
>> -sv

I really don't trust the idea of p2p networks. If a server could cache 
the most downloaded files to memory, then serve them out to a host of 
users at the same time, then it seems that there would be only one need 
for transmitting the file out to multiple users. The server should be 
loaded less, because of only needing to broadcast once for a lot of users,

I think the server side needs a bit more perfecting the transmission of 
files from the server to the clients.

If there is deemed a different protocol to transfer files from server to 
user, then developing a better file transfer system might be a better 
concept.

Jim