Slashdot Mirror

With sendfile () you have an open socket and you tell the kernel to send a file over that socket. No more copying of data and no context-switches.

Sendfile() (and mmap()) are, indeed, the proper interfaces for sending an entire file out. Unfortunately, they aren't particularly well-used — the speed of RAM on modern computers usually makes any wins from funky APIs not worth the effort. It does not help, that various OSes have a completely different call named "sendfile"...

Even Varnish — which most people put in front of their CMS-powered sites — abolished it some time ago (years after advising users to disable it first)... Sad, but true.

With sendfile () you have an open socket and you tell the kernel to send a file over that socket. No more copying of data and no context-switches.

Sendfile() (and mmap()) are, indeed, the proper interfaces for sending an entire file out. Unfortunately, they aren't particularly well-used — the speed of RAM on modern computers usually makes any wins from funky APIs not worth the effort. It does not help, that various OSes have a completely different call named "sendfile"...

Even Varnish — which most people put in front of their CMS-powered sites — abolished it some time ago (years after advising users to disable it first)... Sad, but true.

Why HTTP/2.0 does not seem interesting

Why are we discussing SSL? The guy who develops Varnish has said that SSL is a mess, OpenSSL is a confusing and terrifying mess, and SSL is bad because he doesn't understand the code.

https://www.varnish-cache.org/docs/trunk/phk/ssl.html

First, I have yet to see a SSL library where the source code is not a nightmare.

As I am writing this, the varnish source-code tree contains 82.595 lines of .c and .h files, including JEmalloc (12.236 lines) and Zlib (12.344 lines).

OpenSSL, as imported into FreeBSD, is 340.722 lines of code, nine times larger than the Varnish source code, 27 times larger than each of Zlib or JEmalloc.

This should give you some indication of how insanely complex the canonical implementation of SSL is.

Second, it is not exactly the best source-code in the world. Even if I have no idea what it does, there are many aspect of it that scares me.

Translation: SSL libraries are big and scary, SSL is big and confusing and I have no idea what the hell it does so it's bad.

Ignore this, since it wouldn't help all that much. I'd say that https://www.varnish-cache.org/ can help, but to be honest, if you want to be up, just stick them on Amazon Cloud services or something. They'll have a really hard time getting to that, and you'll leave all the defending and whatnot to be someone elses' job

Adding a reverse cache (such as https://www.varnish-cache.org/ ) in front of your smart web servers for static and mostly static content really helps until they overload your pipe or start hitting dynamic URL's.

You should have setup a Varnish instance in front of it. A bit of a plug for some foss software, but it does work *really* well. In every place I have ever deployed it webpage load, even from spikes like "slashdoting", is never a problem if its configured remotely correctly.

Poul-Henning writes buggy code and has no business writing this sort of article

I think someone should leak this url to them.. 503 errors are not cool, you know? Especially for someone who tries to cultivate an image of skillz and 1337 (and probably bunnies, too)

An alternative could be a Varnish in front, with lets say... 1 hour cache for anon users?

Poul-Henning Kamp of the Varnish about autotools: "Did you call them autocrap tools ? Yes, in fact I did, because they are the worst possible non-solution to a self-inflicted problem." Read more at: http://www.varnish-cache.org/docs/2.1/phk/autocrap.html

Error 503 Service Unavailable

Service Unavailable

Guru Meditation:

XID: 2144369020

Varnish

WTF is "Varnish", and why does a Slashdot 503 link to it?

Here's a good writeup from a developer who definitely expects you to be using swap:
http://varnish-cache.org/wiki/ArchitectNotes

In brief, if you want Varnish to keep a 8GB HTTP cache, then it's going to malloc() 8GB. It expects the kernel to manage swapping whatever parts of that are unused.
He points out that in an app such as Squid, which tries to maintain its own disk cache alongside its RAM cache, the two swap algorithms fight each other in a very counter-productive way.

I think all developers ought to be free to take that same approach. Why should an application programmer write potentially substandard memory management routines when the OS is supposed to be good at it. I agree there will be outliers where the application has special domain knowledge that it can use to manage memory more effectively than the kernel. I suspect there are *very* few such cases.

... the RSS feed for this article fails to load!

Error 503 Service Unavailable

Service Unavailable

Guru Meditation:

XID: 1704629829

Varnish