Hi, I'm interested in adding deltarpm support to yum in the fedora updates infrastructure. There's already a yum plugin written by a redhat developer, so we have something to start from. I have contacted the developer, and he has no problem helping us move along.
Benefits: ======= 1- Clients download faster updates. The README says drpm based update infrastructure can reduce required bandwidth to about 20% on the average 2- Also, from the server side, this should decrease our bandwidth requirements, and free the servers quickly to handle other users
Notes ===== 1- The code currently expects redhat satellite server file system hierarchy, so it will need some cleaning/polishing 2- Currently, the server side stores *all* updates issued, when a client requests updating, the server generates an *appropriate* drpm, which the client can download. This has the disadvantages of storing all updates ever released on the server, and also, running active code on the server might not be welcome by mirrors I imagine. This is a current problem! 3- As a solution to that problem, I am proposing we statically (cron-job) generate drpms only for the newest updates. That way we will be serving the majority of users drpms. We will also get rid of having to generate drpms on the fly. The only thing we loose, is if someone is slow to update his system, he won't be benefiting much from the system. Practically, I think the benefits outweigh the cons. What do you think?
On 1/12/07, Ahmed Kamal email.ahmedkamal@googlemail.com wrote:
Benefits:
1- Clients download faster updates. The README says drpm based update infrastructure can reduce required bandwidth to about 20% on the average 2- Also, from the server side, this should decrease our bandwidth requirements, and free the servers quickly to handle other users
Anything that makes the end user experience better sounds good to me. My questions are, what does it take on our end? And does it require the mirrors to be altered in any way?
-Mike
On Sat, 2007-01-13 at 10:10 -0600, Mike McGrath wrote:
On 1/12/07, Ahmed Kamal email.ahmedkamal@googlemail.com wrote:
Benefits:
1- Clients download faster updates. The README says drpm based update infrastructure can reduce required bandwidth to about 20% on the average 2- Also, from the server side, this should decrease our bandwidth requirements, and free the servers quickly to handle other users
Anything that makes the end user experience better sounds good to me. My questions are, what does it take on our end? And does it require the mirrors to be altered in any way?
Is the 20% for something that applies to the rpms or something that applies directly to the filesystem? I'm not too keen on the filesystem approach although I hear it can really improve the speed. Patching the rpms would be nice... but does require keeping old versions of the rpms on the end-user's disk.
-Toshio
infrastructure@lists.fedoraproject.org
-
Ahmed Kamal -
Mike McGrath -
Toshio Kuratomi