Re: storeGet() -> storeGetPublic() ?

From: Adrian Chadd <adrian@dont-contact.us>
Date: Sat, 6 Jan 2001 07:25:53 +0800

On Fri, Jan 05, 2001, Duane Wessels wrote:
>
>
> On Fri, 5 Jan 2001, Adrian Chadd wrote:
>
> > I am asking because I'm going to have to turn storeGetPublic() into
> > a blocking operation (because it will become the old storeOpen() -
> > it could possibly require a disk operation to get the metadata
> > in before it returns..) and I'd like to kill the rest of the storeGet()
> > calls.
>
> should make ICP nice and speedy.
>

Well, it'll have the opposite effect to start with - since its blocking,
icp requests will take longer to complete. *BUT*, it'll have an added
side-effect - whenever an object is ICPed, it'll need the start of the
object paged into RAM, which will hint the OS to start paging the object
in. So, when you have the metadata read in (so you can actually generate
a StoreEntry) you'll end up having the beginning of an object in, so
HITs will be faster.

(Thats assuming your FS is like reiserfs_raw and keeps the index ondisk..
I imagine that an implementation of the current UFS code will keep
an in-memory index and so these operations will take a few CPU
instructions longer .. :)

Note again that this is only happening on the modio branch for now.
If noone objects I'll port over the patches to squid-HEAD to tidy
things up a little.

ADrian

-- 
Adrian Chadd			"Here's five for the cake, and
<adrian@creative.net.au>	  five to buy a clue."
				    - Ryan, Whatever it Takes
Received on Fri Jan 05 2001 - 16:26:05 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:13:10 MST