Freebsd Theme Song

[Ted Mittelstaedt]

Michael,

  Fundamentally, here's the problem Danial is claiming exists:

it takes a certain amount of time to get the packet clocked in
from the network into the ethernet receiver.  This is hardware
dependent and cannot be changed.

It takes a certain amount of time to get the packet out of
the hardware in the ethernet card into main ram, this also
hardware dependent and cannot be changed. (unless the device
driver is terribly inefficient, which we will assume it's not)

Once in main ram, the information in the packet has to go through
a number of code statements.  The more code statements the
longer the information in the packet is sitting around in
the FreeBSD system's memory.

It then takes a certain amount of time to get the information
out of main memory into the other sending ethernet nic's buffers,

and it takes time to get it out of the sending nic back to the
wire.

Danial is claiming the slowness is in the main ram section of
things, not in the ethernet driver code.

polling makes the ethernet driver more efficient at high data
rates, but it does nothing for the speed of processing within
the TCPIP stack itself.  At low data rates polling is less
efficient than the interrupt method.  And unless the nic driver
is terribly inefficient to start with, the time it adds to the
packet path in the system is minor compared to the time spent
in the TCP/IP stack.

Ted

>-----Original Message-----
>From: owner-freebsd-questions at freebsd.org
>[mailto:owner-freebsd-questions at freebsd.org]On Behalf Of Michael Vince
>Sent: Monday, December 12, 2005 3:26 AM
>To: danial_thom at yahoo.com
>Cc: freebsd-questions at freebsd.org; Kris Kennaway
>Subject: Re: Freebsd Theme Song
>
>
>What about just turning on Polling?
>I have polling turned out for a router and all I get is gigabit
>performance.
>I have tested it with a wire variety of tests from basic fetch tests
>from a FreeBSD client box via a FreeBSD  router (with polling) to
>another FreeBSD box and all I got was gigabit performance in
>either 'ab'
>tests (would always gave 114megabytes/sec, or even just doing a 'fetch'
>of a single 1gig file I could get up to around 90 megabytes/sec
>which is
>largely file system read performance limited over network performance
>limited.
>
>Its the same with my Samba server sure without polling I get quite
>ordinary network performance but when I turn on polling its appears to
>be limited by the gigabit cable quality setup and the switch
>and quality
>of hardware like using Intel em gigabit ethernet devices.
>
>I agree that networking performance is really important and I do agree
>that FreeBSD out of the box doesn't perform as well as it could
>in those
>areas but there are some solutions for it that fill the gaps
>for all the
>situations I have faced, I plan to use them for as long as I need till
>things like interrupt latencies can be over come.
>
>People should enjoy FreeBSD for what it is, something thats not holding
>you back anywhere, there are countless examples.
>There is no one trying to design a system to squeeze money out of you,
>their not trying to force you to buy a rpm up2date system.
>They aren't holding you down with package choices such as being
>stuck on
>a old version of apache 2.0.x that just gets 'security' patched and
>never gets a version increment so you miss out of performance
>improvements of a particualy module of the stable Apache 2.0.x series,
>just so they can try and sell you a new version of CD so you can do a
>binary upgrade, the list can go forever.
>These systems are designed to control you and at the same time limit
>your possibilities because they 'fear' loosing that control of you,
>FreeBSD has no 'fear' riddled/limiting motivations because it has no
>evil intentions, and just like real freedom look at your choices you
>get. Fear is the path to the darkside.
>
>Alright I am going off topic, what I am trying to say is I
>think you are
>entiled to say what you like, I have sometimes thought in somewhat
>similar ways, but I also believe you should try and be happy with what
>you get from FreeBSD and if you really want things to move on then one
>of the best things that can be done is either raising funds for
>developers to work on it or providing code your self.
>
>Mike
>
>Danial Thom wrote:
>
>>Also, since you don't see to understand the test,
>>bridging is not routing. Its a rote function of
>>moving packets from one interface to another with
>>very little overhead. Its purely interrupt
>>driven, so the kernel's latencies in processing
>>interrupts is well exercised. Its a good test
>>because, unlike crap like netperf, it doesn't
>>involve sockets or any userland tasks. I know
>>you're not a real engineer Kris, so I don't
>>expect you to understand, but you also aren't
>>qualified to discredit the test, since you don't
>>know a damn thing about testing.
>>
>>I know you enjoy being the one-eyed man in the
>>land of the blind on this list Kris, But I doubt
>>people are stupid enough to buy into your
>>continued propaganda. There isn't one credible
>>test that shows that FreeBSD MP is worth any
>>consideration as a good performer, so it seems
>>doubtful that anyone with half a brain thinks it
>>is.
>>
>>Everything today is networking. What good is a
>>fast filesystem if it sits on a klunky kernel or
>>slow networking system? Who's going to build a
>>big honking MP server if is can't handle more
>>network traffic than a good UP system?
>>
>>Do you have a volkwagon engine in your Porche,
>>Kris? The problem with Kris is that he thinks
>>that if his car has a really cool radio that
>>people will buy it, even those its slow as shit.
>>That may be fine for the kind of guys that hang
>>out on the freebsd-questions list, or for little
>>old ladies. But its not "fine" with the kind of
>>people that used to rely on FreeBSD for serious
>>networking tasks.
>>
>>Kris is just a PR front man for a "team" of
>>developers that is lost. Their "theory" on how to
>>build a better mousetrap for MP is completely
>>wrong, and now they're going to try something
>>else, using the entire FreeBSD community as
>>guinea pigs. First 5.4 was the answer. Then 6.0.
>>Now it looks like 6.0 sucks too. Its a damn
>>shame.
>>
>>DT
>>
>
>_______________________________________________
>freebsd-questions at freebsd.org mailing list
>http://lists.freebsd.org/mailman/listinfo/freebsd-questions
>To unsubscribe, send any mail to
>"freebsd-questions-unsubscribe at freebsd.org"
>
>--
>No virus found in this incoming message.
>Checked by AVG Free Edition.
>Version: 7.1.371 / Virus Database: 267.13.13/197 - Release
>Date: 12/9/2005
>