On Wed, 2002-09-25 at 23:32, Steve Snyder wrote:
>
> In the Squid v2.5 release notes it is stated that pipeline_prefetch "Now
> defaults to off for bandwidth management and access logging reasons".
> What are those reasons? My understanding is that pipelining just
> requests multiple objects concurrently rather than sequentially. Can
> someone go into a little more detail on this?
Pipelining is still sequential. The additional requests are sent on the
same connection, to reduce latency. However, this means that you get
multiple objects downloading at once, which interacts with delay pools,
and if, for some reason, the first request causes the connection to
shutdown, the pipelined request is cancelled, thus wasting the data that
was brought down, if it hadn't completed before the first request
shutdown the connection.
Access logging I'll leave to Henrik to answer. I don't recall what the
specific issue there was.
Rob
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:10:24 MST