John created TS-3662:
------------------------
Summary: Caching range requests plugin.
Key: TS-3662
URL: https://issues.apache.org/jira/browse/TS-3662
Project: Traffic Server
Issue Type: Bug
Components: Plugins
Reporter: John
We are using ATS to cache very large binary files 60gb + for a customer. The
customers user client software is used to fetch these files using HTTP range
requests in a predictable manner ie, all clients request the same byte ranges.
In any event, if we were to cache the large 60gb file obtained from the origin,
the ATS would save it to a single stripe or disk drive using the background
fetch plugin. When we tried this method, we discovered that the thousands of
range requests to a single disk drive caused a significant performance issue,
high I/O wait and load average on the host running ATS.
The plugin I have written, see the attached patch, transforms the range
requests at various points in the HTTP flow so that the ATS may cache the range
requests as individual objects by modifying the the request URL, removing and
restoring the range request headers where appropriate. This allows the
individual range requests to be stored and retrieved as individual objects that
are spread over multiple disk drives. See the source as it is well commented.
We have this plugin in production within our CDN and have seen a dramatic
improvement for delivery of our customers files using range requests.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
