Conditional backend requests on repository branch experimental-ims

Geoff Simmons geoff at uplex.de
Sun Jun 19 20:30:37 CEST 2011


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 6/9/11 2:13 AM, Paulo Paracatu wrote:
> 
> Yesterday I finally had the time to download this and make the RPM to
> install in my caches servers.

Thanks for trying it out, and sorry for the late response. The
experimental-ims branch in the repository is now up to date with the
current master branch, including the 3.0 release.

> *      252531       416.87       344.05 fetch_304 - Fetch no body (304)*
> 
> Ah, before the patch I was hitting much more RX than now, so I guess it
> is really working and helping.
> 
> If you have any tips, I'd be glad to test.

Since a few people have asked how to make sure that the conditional
requests are working, I added a section about verification to the Wiki page:

http://www.varnish-cache.org/trac/wiki/BackendConditionalRequests#Verification

Non-zero values of fetch_304 definitely indicate that Varnish has sent
conditional requests. Watching varnishlog is the best way to know
exactly what's going on.

As for testing: any exposure to a live web site will be very helpful.
All of my testing has been with varnishtest test cases and synthetic
performance tests. I haven't been able to break it for a while using
load generators for heavy load and high concurrency, but nothing beats
the real thing.

I haven't tested it with ESI, which has been improved for 3.0. Varnish
with the IMS feature should work smoothly with conditional requests for
ESI includes, but I'm not familiar with ESI and haven't confirmed that.
(I have confirmed that conditional requests work correctly with objects
that are compressed in the cache.)

The feature introduces some additional copying -- Varnish always creates
a new object from a backend fetch, and the IMS function copies the body
and headers from the stale object to the new object when a 304 is
returned. It would be good to test it with large objects to see if there
are signs of performance loss.

Thanks again for the feedback!


Geoff
- -- 
UPLEX Systemoptimierung
Schwanenwik 24
22087 Hamburg
http://uplex.de/
Mob: +49-176-63690917
-----BEGIN PGP SIGNATURE-----
Version: GnuPG/MacGPG2 v2.0.14 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
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=oGO8
-----END PGP SIGNATURE-----




More information about the varnish-misc mailing list