This work aims at providing optimized request forwarding for accelerator
setups by collapsing multiple requests for the same object into one backend
server request.
Table of contents
- Documentation
- Progress
- Configuration
- Internal design
- Known issues and shortcomings
In accelerator setups it is desireable if the number of connections to
the backend web servers is minimized. However, Squid being designed primarily
for forward proxy operation does not take this into consideration. As a
result there may be storms of requests to the backendserver if a very
frequently accessed object expires from the cache or a new very frequently
accessed object is added.
To remedy this situation this patch adds a new tuning knob to squid.conf,
making Squid delay further requests while a cache revalidation or cache miss
is being resolved. This sacrifices general proxy latency in favor for
accelerator performance and thus should not be enabled unless you are
running an accelerator.
In addition an option to shortcut the cache revalidation of frequently
accessed objects is added, making further requests immediately return as
a cache hit while a cache revalidation is pending. This may temporarily
give slightly stale information to the clients, but at the same time allows
for optimal response time while a frequently accessed object is being
revalidated. This too is an optimization only intended for accelerators,
and only for accelerators where minimizing request latency is morer
important than freshness.
- 2006-09-27 Bug #1780: collapsed_forwarding and Vary
- Need to deal with Vary when collapsing multiple requests for the same URL.
- 2006-05-18 Merged into Squid-2.6
- The collapsed forwarding support has been merged into the upcoming Squid-2.6 release.
- 2003-07-28
- Final release for Squid-2.5. Fixes problems with stuck objects,
ims related bug fixes, window based refreshes and more.
- 2003-06-21 First public release
- First public release as a patch to Squid-2.5
- 2003-06-18 Final delivery
- Final delivery of the changes sent to customer
- 2003-06-15 Initial delivery
- Initial delivery of the changes sent to customer
- 2003-05-18 Project signed
- The project was initiated
- collapsed_forwarding on/off (default off)
- This option enables collaps of multiple requests for the same URI
to be processed as one request. Normally disabled to avoid corner
cases with hung requests, but there can be large benefit from
enabling this in accelerator setups where the web servers are
reliable.
- refresh_stale_window interval (default 0)
- This option decreases latency on collapsed forwarding by initiating
a revalidation request some time before the object becomes stale.
This avoid having more than one client wait for the revalidation
to finish.
This effect is acheived by the following primary changes
- On cahce misses the new object is made immediately public, allowing
new requests to attach to the pendign request. If the object is later
found to be private then the attached requests will all detach and initiate
new requests.
- On cache revalidations entry->mem_obj->ims_entry is set to the
StoreEntry of the IMS query, allowing additional requests to optionally attach
to the same backend IMS query.
- To avoid objects which for one or another reason gets stuck and does not receive
a reply in a timely fashion entry->mem_obj->refresh_timestamp is used to
avoid objects which has been pending for IMS reply more than 30 seconds.
- The optional refresh_stale_window also uses the same refresh_timestamp field
to keep track of if a revalidation has been initiated or not.
The stale_hit effect uses the same mechanism by disregarding that the
object is stale if there is already a cache revalidation running (ims_entry set).
- None. Might even be suitable for the general Internet proxy situation, not only reverse proxies.
$Id: index.html,v 1.3 2006/09/27 11:25:49 hno Exp $