1.6 Million IP Connections on FreeBSD

An anonymous reader writes "FreeBSD developer Terry Lambert, in a recent posting to the 'freebsd-hackers' mailing list, mentioned that he'd tuned a FreeBSD 4.4 box with 4GB of RAM to achieve 1,603,127 simultaneous IP connections, and goes on to say: 'As far as I know, I hold the single machine connection record for an x86 box.' This is an impressive achievement any way you look at it (though it begs the question of whether or not the box had any resources left to actually do anything with those connections...), and it speaks well of both FreeBSD's capabilities and Terry's skills and knowledge. I'm curious, though, if anyone has approached, matched, or exceeded that number elsewhere?"

Comments

Alternative Headline:

[leviramsey]

*BSD handles 1.6 million connections without dying.

sure

[pizza_milkshake]

my webhosting box does twice that during peak hours, but then i'm hosting free porn so it doesn't count ;)

Prove it...

[aridhol]

Post the address of that box here. We'll give it a real stress test.

More Interesting

[mnmn]

What has been tested is simply the number of concurrent connections. More practical would be simple retrieving of say 1kb data from a database and printing it out on a very simple HTML, and checking the maximum number of THESE connections. In effect trying to really httpblast DDoS style the FreeBSD with sheer number of connections. The box will have to be massive with 4GB RAM at least (we're testing OS here not hardware) and the connection maybe (multiple?) gigabit ethernet. The result would theoretically be lower than 1.6 million but we need to show FreeBSD can scale in practical tests like these. Results from a test like that will have the power to change vendors' minds from trying to run IIS and MS SQL for a high volume site.

Re:More Interesting

[hfastedge]

theres no point in the mysql test on top of the ip test, it then becomes even moreso a pure kernel test (eg scheduler/vm....), plenty of which have been done.

But i do think that serving out 1k of html would make the test a bit more solid.

Id be interested, given this narrow field thats being tested on how linux would hold up under the same tweaking, and what tweaking exactly that would require as compared to the freebsd (yes its probably trivial, but im just curious).

Re:More Interesting

[wabb1t]

Great! There were rumours that FreeBSD was still being used at Hotmail, and this seems like possible proof.

Now if we could only see some actual hotmail pages served by that machine...

Re:More Interesting

[DotComVictim]

The test you propose is not a more interesting test. It is simply a different test.

The original test is designed to stress the theoretical maximum number of connections, which exercises the network stack, in particular the pcb hashing mechanisms and multiple IP address handling.

The test you propose is a real world scalability test, which has a much different purpose.

FreeBSD is dying

[mcgroarty]

...to take on heavy server loads.

Re:Bahhh.

[tigga]

This is just something else Apple will steal from the BSD'ers.

You can't steal anything that already free.

Re:"Connections" -- ?

[cperciva]

You evidently didn't RTFA carefully enough -- the subject line ("max simultaneous TCP connections") should have been a giveaway.

Re:Bahhh.

[Arandir]

This is just something else Apple will steal from the BSD'ers.

There once has a fabulous apple tree. No matter how many apples one would take from it, there were just as many as before! When this was heard by the villagers they all rushed to the apple tree and took apples. But no matter how many they took, there were just as many apples as before. But some of them came and took apples and locked them within a chest, so that none could steal them. And they laughed at the other villagers, saying, "Look, they do not protect their apples. Surely a thief will come and steal them."

64-bit version

[yancey]

I wonder if the 64-bit version of FreeBSD would be able to improve upon this, since it can access more memory.