Difference between revisions of "Winter 2011 Handout - CreateRepo Performance Improvements"
(→Challenges) |
(→Approach) |
||
| Line 30: | Line 30: | ||
| − | Currently I am doing load testing on Hong Kong and Ireland to see which part of the hardware on the machine heavily hit by the createrepo process. | + | Currently I am doing load testing on Hong Kong and Ireland to see which part of the hardware on the machine heavily hit by the createrepo process. To see which put of the machine are heavily use durning a createrepo I create some dumpy repo to see which resource are heavily. tool that I am using to load tests are sar, iotop, and vmstat. |
Revision as of 21:53, 12 April 2011
Contents
Status
This page status is: DRAFT.
CreateRepo Performance Improvements
Participants
Derrick Boddie
Introduction
The Fedora-ARM build farm consists of the koji-hub/koji-web system (on HongKong, an x86_64 machine) and a group of ARM builders. HongKong also handles the CreateRepo tasks, which heavily load that machine. It might make sense to move those tasks to another machine, but doing so makes the CreateRepo jobs take a lot longer (15 minutes on HongKong vs. 55 minutes on another x86_64 server). This is presumably due to the overhead of sharing files between HongKong and the other server over NFS on the 100 Mbps Seneca network.
Approach
This project involves figuring out how to run the CreateRepo jobs more quickly. Possible solutions include a 1 Gbps LAN, a redistribution of the file storage or a change to a different file storage technology, or optimizing the CreateRepo tasks on HongKong for best speed.
To improve performance of the createrepo I set up a 1 Gbps LAN between Hong Kong and Ireland. We hope that the faster network connection between Hong Kong and Ireland would bring better createrepo performance.
Currently I am doing load testing on Hong Kong and Ireland to see which part of the hardware on the machine heavily hit by the createrepo process. To see which put of the machine are heavily use durning a createrepo I create some dumpy repo to see which resource are heavily. tool that I am using to load tests are sar, iotop, and vmstat.
