On Tue, Dec 04, 2001 at 12:48:45PM +0100, Nicolas Souchu wrote:
> Question, what is RTT? The subject seems interesting but without the
> background... :)
Round Trip Time. The time it takes for a segment to travel to the
destination, be processed, and an ack returned.
--
Leo Bicknell - [
On Fri, Nov 30, 2001 at 12:58:39PM -0500, Leo Bicknell wrote:
>
> Since the topic has come up again, I'll provide some graphs, and
> go back to my suggestion to see if it gets some traction this time
> around.
>
> http://www.ufp.org/~bicknell/fbsdtcp.png
>
> This graph shows the theoretical max
Leo Bicknell wrote:
>
> On Fri, Nov 30, 2001 at 10:29:28AM -0800, Luigi Rizzo wrote:
> > It is not a big deal to move the default to 32 or 64k, and I'd
> > vote for that, but if a sysadmin is unable to have a look at this,
> > then the problem is in the sysadmin, not in FreeBSD!
>
> I disagree,
:I think I tried this patch, and found some problems with it. As
:I recall the problems were with extremely high bandwidth connections
:(eg, I have two machines that can move 100Mbps FDX across country
:(70ms latency), and when I tried the patch with that case performance
:was "bad", in the sens
On Fri, Nov 30, 2001 at 05:48:16PM -0800, Luigi Rizzo wrote:
> On Fri, Nov 30, 2001 at 08:39:05PM -0500, Leo Bicknell wrote:
> > Note that if we implement a 'fair share' buffering scheme we would
> > never get a failure, which would be a good thing. Unfortuantely
> > fair share is relatively comp
On Fri, Nov 30, 2001 at 08:39:05PM -0500, Leo Bicknell wrote:
> On Fri, Nov 30, 2001 at 05:30:33PM -0800, Luigi Rizzo wrote:
> > I just committed to current (and soon to stable) some code to log
> > _failures_ in mbuf allocations, but that is only meant as an aid
> > to remove worse code in the d
On Fri, Nov 30, 2001 at 05:30:33PM -0800, Luigi Rizzo wrote:
> I just committed to current (and soon to stable) some code to log
> _failures_ in mbuf allocations, but that is only meant as an aid
> to remove worse code in the drivers.
Note that if we implement a 'fair share' buffering scheme we
On Fri, Nov 30, 2001 at 05:19:18PM -0500, Mike Silbersack wrote:
>
> On Fri, 30 Nov 2001, Leo Bicknell wrote:
>
> > * The logging at 90% usage should be investigated. I can probably
...
> Luigi, Jonathan and I had already been discussing this idea before this
> this thread even started. If you
On Fri, Nov 30, 2001 at 12:59:53PM -0800, Matthew Dillon wrote:
> The transmit side requires more thought. I did write that patch, and
> it does work, but it's too messy for my tastes. I would personally
> much rather rewrite it to (A) fix the RTT stored in the route tables
> an
Alfred Perlstein wrote:
> Hmm, well the GENERIC default is some mathematical operation on
> maxusers. We really ought to make this scale as a default relative
> to the amount of ram in the system, rather than some low hardcoded
> value. NetBSD has some stuff for this in their buffercache sizing
* Leo Bicknell <[EMAIL PROTECTED]> [011130 16:14] wrote:
>
> First off, apologies to Luigi, I was shooting off my mouth.
Understandable, it's easy to get heated about an issue when
it weighs so much in ones mind. I've done the same on several
quite memorable occasions.
> Second off:
>
> On Fr
On Fri, 30 Nov 2001, Leo Bicknell wrote:
> * The logging at 90% usage should be investigated. I can probably
> generate patches for that over the weekend, provided I can find
> a good way to rate limit them.
Luigi, Jonathan and I had already been discussing this idea before this
this threa
On Fri, Nov 30, 2001 at 05:14:18PM -0500, Leo Bicknell wrote:
>
> First off, apologies to Luigi, I was shooting off my mouth.
no problem, and no need for apologies :)
cheers
luigi
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of t
First off, apologies to Luigi, I was shooting off my mouth.
Second off:
On Fri, Nov 30, 2001 at 01:50:42PM -0600, Alfred Perlstein wrote:
> I was about to set the default in -stable to Leo's suggested values,
> it seems that -current already has the delta he wants in it,
> my question is, was a
> > The question that immediately comes to mind is, why not simply use
> > as big a value as possible? The problem comes down to buffering
> > the data, and busy servers may have to buffer a lot of data. Having
> > a 1 meg window size may have you buffer 1 meg per connection. Note
> > that
:I don't think anyone's doubting the importance of larger windows; it's
:just that we can't do much increasing until they're dynamic.
:
:That being said, Matt did post a patch which implements socket buffer
:autoscaling a few months back. I've been meaning to review it, but
:haven't had the time.
On Fri, 30 Nov 2001, Leo Bicknell wrote:
> Since the topic has come up again, I'll provide some graphs, and
> go back to my suggestion to see if it gets some traction this time
> around.
>
> http://www.ufp.org/~bicknell/fbsdtcp.png
I don't think anyone's doubting the importance of larger window
* Alfred Perlstein <[EMAIL PROTECTED]> [011130 13:51] wrote:
>
> I was about to set the default in -stable to Leo's suggested values,
> it seems that -current already has the delta he wants in it,
> my question is, was anything else changed along the lines of the
> number of nmbclusters allocated
* Luigi Rizzo <[EMAIL PROTECTED]> [011130 13:26] wrote:
> On Fri, Nov 30, 2001 at 02:11:00PM -0500, Leo Bicknell wrote:
> ...
> > * Many people use FreeBSD as a desktop OS. Think the same people
> > who use Win98, but only slightly smarter. These people are
> > 'sysadmins' only in the sense
: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]]On Behalf Of Leo Bicknell
> Sent: Friday, November 30, 2001 11:11 AM
> To: Luigi Rizzo
> Cc: [EMAIL PROTECTED]
> Subject: Re: TCP Performance Graphs
>
>
> On Fri, Nov 30, 2001 at 10:29:28AM -0800, Luigi Rizzo wrote:
> &
On Fri, Nov 30, 2001 at 02:11:00PM -0500, Leo Bicknell wrote:
...
> * Many people use FreeBSD as a desktop OS. Think the same people
> who use Win98, but only slightly smarter. These people are
> 'sysadmins' only in the sense that they have a root password.
> When FreeBSD can't fill their
On Fri, Nov 30, 2001 at 10:29:28AM -0800, Luigi Rizzo wrote:
> It is not a big deal to move the default to 32 or 64k, and I'd
> vote for that, but if a sysadmin is unable to have a look at this,
> then the problem is in the sysadmin, not in FreeBSD!
I disagree, on two points:
* Many people use F
On Fri, Nov 30, 2001 at 01:47:41PM -0500, Andrew Gallatin wrote:
> I thought that I heard a few months ago that Matt Dillon was looking
> at ways to dynamically size tcp windows from within the kernel. Maybe
> I'm on crack.
He is. It is very good work that I wish I could spend more time
helping
Leo Bicknell writes:
> The question that immediately comes to mind is, why not simply use
> as big a value as possible? The problem comes down to buffering
> the data, and busy servers may have to buffer a lot of data. Having
> a 1 meg window size may have you buffer 1 meg per connection.
The default window size (controlled by the socket buffer size) can
be globally modified using sysctl variables:
net.inet.tcp.sendspace: 16384
net.inet.tcp.recvspace: 16384
As you mention, changing this (and other things such as the amount
of mbufs/clusters, etc.etc.) must be done
25 matches
Mail list logo
