didn't we all already know this? http://financial.washingtonpost.com/industry_list.asp?mode=news&doc_id=BW20010606BW2321&ticker=LU -craig
Date: Wed, 6 Jun 2001 10:54:56 -0400 From: Craig A. Haney <craig@seamless.kludge.net>
didn't we all already know this?
http://financial.washingtonpost.com/industry_list.asp?mode=news&doc_id=BW20010606BW2321&ticker=LU
Journalists. *snort* I know that _I_ did, and I'm no backbone engineer... it's a matter of statistics. Anybody else remember a recent thread arguing that QoS should be kept at the edge for IP?
-craig
Eddy --------------------------------------------------------------------------- Brotsman & Dreger, Inc. EverQuick Internet Division Phone: +1 (316) 794-8922 Wichita/(Inter)national Phone: +1 (785) 865-5885 Lawrence --------------------------------------------------------------------------- Date: Mon, 21 May 2001 11:23:58 +0000 (GMT) From: A Trap <blacklist@brics.com> To: blacklist@brics.com Subject: Please ignore this portion of my mail signature. These last few lines are a trap for address-harvesting spambots. Do NOT send mail to <blacklist@brics.com>, or you are likely to be blocked.
On Wed, 6 Jun 2001, Craig A. Haney wrote:
didn't we all already know this?
[ Their surprising discovery - that traffic on heavily loaded, high-capacity network links is unexpectedly regular ] These characteristics had been pointed out by Bill Barns and MO in the core of a promising local ISP. The phrase used was something like "city pair macro flows are long lived and stable." It is good to get some formal research into this, backing up emperical data. /vijay
The Bell Labs guys only used "hundreds of gigabytes" of "real" packet data, which if I am not mistaken is only a few minutes of traffic on any high speed link. The more concerning issue with the article (I found) was that they "simulated real data". What does this mean? How can you simulate something you are trying to study without prejudicing your results? But yes, I think I have heard/seen "macro flows are long lived and stable" before, but didn't know who to credit it to. Deepak Jain AiNET -----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of Vijay Gill Sent: Wednesday, June 06, 2001 11:40 AM To: nanog@merit.edu Subject: Re: Internet Traffic Discovery? On Wed, 6 Jun 2001, Craig A. Haney wrote:
didn't we all already know this?
[ Their surprising discovery - that traffic on heavily loaded, high-capacity network links is unexpectedly regular ] These characteristics had been pointed out by Bill Barns and MO in the core of a promising local ISP. The phrase used was something like "city pair macro flows are long lived and stable." It is good to get some formal research into this, backing up emperical data. /vijay
At 10:54 AM -0400 6/6/01, Craig A. Haney wrote:
didn't we all already know this?
I thought that we "knew" that, according to self-similarity, burstiness would be seen at all levels of aggregation. I wonder if scaling rules apply as the aggregation increases (backbones). That is, at high levels of aggregation, would we have to look at smaller time intervals to see the burstiness? Don't know if the Ball Labs folk did that. --Steve --
On Wed, Jun 06, 2001 at 11:40:41AM -0400, Steve Goldstein wrote:
At 10:54 AM -0400 6/6/01, Craig A. Haney wrote:
didn't we all already know this?
I thought that we "knew" that, according to self-similarity, burstiness would be seen at all levels of aggregation.
I believe Dennis Ferguson brought this assumption under reasonable doubt using data from MCI a while back. -dorian
At 11:54 AM -0400 6/6/01, Dorian Kim wrote:
...
I thought that we "knew" that, according to self-similarity, burstiness would be seen at all levels of aggregation.
I believe Dennis Ferguson brought this assumption under reasonable doubt using data from MCI a while back.
Very possible, Dorian. I have not been following very carefully. Thanks for pointing this out. BTW. please note my use of " " around "knew." --SteveG --
On Wed, Jun 06, 2001 at 11:54:42AM -0400, Dorian Kim wrote: On Wed, Jun 06, 2001 at 11:40:41AM -0400, Steve Goldstein wrote:
At 10:54 AM -0400 6/6/01, Craig A. Haney wrote:
didn't we all already know this?
I thought that we "knew" that, according to self-similarity, burstiness would be seen at all levels of aggregation.
I believe Dennis Ferguson brought this assumption under reasonable doubt using data from MCI a while back. to quote his operationally spiced eloquence apr 1997 // (and if I hear `self-similarity' one more time I'm going to barf) //
OK, sorry. I had not kept up with the times. --SteveG At 3:00 PM -0700 6/6/01, k claffy wrote:
...
I believe Dennis Ferguson brought this assumption under reasonable doubt using data from MCI a while back.
to quote his operationally spiced eloquence apr 1997
// (and if I hear `self-similarity' one more time I'm going to barf) //
--
On Wed, Jun 06, 2001 at 11:40:41AM -0400, Steve Goldstein wrote:
At 10:54 AM -0400 6/6/01, Craig A. Haney wrote:
didn't we all already know this?
I thought that we "knew" that, according to self-similarity, burstiness would be seen at all levels of aggregation.
I wonder if scaling rules apply as the aggregation increases (backbones). That is, at high levels of aggregation, would we have to look at smaller time intervals to see the burstiness? Don't know if the Ball Labs folk did that.
also interesting would be experimentation in determining what level of route oscillation breaks this axiom. is the stability of macroscopic flows retained for all but macroscopic route oscillation? does it remain even for macroscopic route oscillation? thinking further...is it reasonable to believe that macroscopically internet routing is static? -- Sam Thomas Geek Mercenary
Steve, There are a few things that I see make the Bell Labs study dissimilar from the backbone core. The Bell Labs study looked at LANs and systems with modest levels of aggregation compared to an OC12+ backbone link. The traffic mix for the Bell Labs study included a large amount of NFS traffic, which is high data volume UDP traffic with little flow control. I don't think you'll find much of that on the intercity links, relatively. I also expect the RTT times involved with the TCP traffic on the Bell Labs LAN were much smaller than the norm for intercity links, causing the slow start ramps to be much longer in the later. These things make it reasonable that they would see different results than seen on an OCn link. jerry
participants (9)
-
Craig A. Haney -
Deepak Jain -
Dorian Kim -
E.B. Dreger -
Jerry Scharf -
k claffy -
Sam Thomas -
Steve Goldstein -
Vijay Gill