Their earlier paper - http://citeseer.nj.nec.com/tangmunarunkit01impact.html describes the methodology for generating AS overlay maps.
AS level topology was constructed mainly from the oregon route server bgp dumps.
Anybody can get them from telnet://route-views.oregon-ix.net/ - use "show ip bgp" command. The archives are available at http://archive.routeviews.org/
But, as BGP reveals only the AS preferences instead of actual physical AS connections, the info from "routeviews" is always incomplete. So, other sources like "Internet Routing Registries" like RADB, RIPE (the European registry) are used to fill in the gaps. ( For example http://www.radb.net/cgi-bin/radb/whois.cgi?obj=128.125.19.146 gives the IRR record for USC web server; http://www.radb.net/cgi-bin/radb/whois.cgi?obj=AS226 gives the record for the corresponding AS)
As providing routing policies in IRRs is optional, some ISPs don't do that. Internet "looking glass" sites like http://www.traceroute.org can help providing more details.
To answer your question, I think the difference is that "actual" AS map was obtained directly from BGP dumps like the Oregon route server (which can be augmented by IRRs & looking glasses); http://topology.eecs.umich.edu/archive/asgraph.tar.gz is one such dataset; the "overlay" maps are drawn on the top of the router-level map (obtained by their mercator tool) using the same info from Oregon+RADB. They built the overlay map to get the actual inter-AS router links. The reason they compared the macroscopic properties of the actual & overlay maps is that they used approximation techniques (like collapsing the disjoint nodes of same AS to the nearest AS) in the overlay constructions.
Someone please correct me if I'm wrong.
-Ramesh
----- Original Message -----
From: paavany jayanty <pjayanty@usc.edu>
Date: Sunday, February 2, 2003 4:11 am
Subject: [Csci551-talk] [Q] Internet Path Inflation Due...
> Hi,
>
> In the Internet Path Inflation .... paper there is a mention of
> how overlay AS maps are obtained.
>
> I cannot quite get the difference between the "actual AS map" and
> the AS overlay map.
>
> How are the actual AS maps obtained?
>
>
>
>
From rmgandhi@usc.edu Mon Feb 3 21:57:16 2003
From: rmgandhi@usc.edu (ravi gandhi)
Date: Mon, 03 Feb 2003 13:57:16 -0800
Subject: [Csci551-talk] Regd. Q.3.
Message-ID:
Hii,
I wud like to know how can I test my program on an Intel Machine (little Endian).
Thanks,
Ravi M. Gandhi
MS(C.S.), University of Southern California
710, West 27th Street,
Apartment No.4,
Los Angeles,
CA 90007
Phone: 1-213-749-8983
webpage: http://www-scf.usc.edu/~rmgandhi/
From johnh@ISI.EDU Mon Feb 3 22:08:02 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Mon, 03 Feb 2003 14:08:02 -0800
Subject: [Csci551-talk] "hard" & "soft" states: more pointers?
In-Reply-To: <000001c2c9b4$f7db5d00$0201a8c0@Intrepid>
Message-ID: <200302032208.h13M828v019270@dash.isi.edu>
On Fri, 31 Jan 2003 21:44:20 PST, Ramesh Sarangarajan wrote:
>As I tried to get a clear definition of ühardı & üsoftı states (we had
>opposing views expressed in todayùs class about whether BGP maintains hard
>or soft state), I found this:
>
>
>
>http://users.exis.net/~jnc/tech/hard_soft.html
>
>
>
>Can someone provide me pointers to other possible definitions?
>
>
>
>-Ramesh
>
>
>
>
That web page starts to pull apart some of the implications behind
hard and soft state. We'll talk more about this in class when we get
to reliable multicast protocols.
I mostly agree with Chiappa's web page, except for his statement that
soft-state implies that the user is entirely repsponsible for its
maintence. That is not true with many protocols that most people
would agree are soft state, for example flood-and-prune multicast
routing.
-John Heidemann
From johnh@ISI.EDU Mon Feb 3 22:32:57 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Mon, 03 Feb 2003 14:32:57 -0800
Subject: [Csci551-talk] [Q] Is synchronized network traffic bad?
In-Reply-To: <3E3C677A.1020304@usc.edu>
Message-ID: <200302032232.h13MWvQL019454@dash.isi.edu>
On Sat, 01 Feb 2003 16:34:02 PST, JongAm Park wrote:
>I have a question about the synchronized network traffic.
>
>The Floyd's paper, The Synchronization of Periodic Routing Messages,
>makes me think that syncronized traffic is bad and should be avoided.
>Could someone explain why it's something to be avoided?
>The property of synchronization seems to me that it's quite natural.
>
>
This is a very good question. Often people state that something is
bad without stating why. Common examples include synchronization and
oscillation.
As an example of a negative effect of synchronization, consider 5
sources sending at a rate of 1 pkt/s each to a shared queue with 4
buffers. The link drains at a rate of 5 pkts/s. If they all send at
the same time, one packet must always be dropped. If the spread out
their tranmissions over at least 0.2 s (=1/5 s) then no packets are
ever lost.
This doesn't answer why routing synchronization (the Floyd paper)
would be bad. That's a good think to think about more.
-John Heidemann
From johnh@ISI.EDU Tue Feb 4 00:10:53 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Mon, 03 Feb 2003 16:10:53 -0800
Subject: [Csci551-talk] [Q] Internet Path Inflation Due...
In-Reply-To:
Message-ID: <200302040010.h140Arxs020492@dash.isi.edu>
On Sun, 02 Feb 2003 04:11:58 PST, paavany jayanty wrote:
>Hi,
>
>In the Internet Path Inflation .... paper there is a mention of how overlay AS maps are obtained.
>
>I cannot quite get the difference between the "actual AS map" and the AS overlay map.
>
>How are the actual AS maps obtained?
>
I believe the paper refers to "overlay AS map" by comparision to
"router-level map", not in relationship to "actual AS map" (which is
not mentioned, to my knowledge).
Thus the "overlay" part refers to overlaying ASes onto routers and
links.
(See also the references in [Tangmunarunkit01a] and ramesh
sarangaraj's pointers for some reasons the overlay AS map is not
necessarily exactly correct.
-John Heidemann
From sarangar@usc.edu Tue Feb 4 00:55:34 2003
From: sarangar@usc.edu (ramesh sarangarajan)
Date: Mon, 03 Feb 2003 16:55:34 -0800
Subject: [Csci551-talk] [Q] Internet Path Inflation Due...
Message-ID:
> >I cannot quite get the difference between the "actual AS map" and
> the AS overlay map.
>
> I believe the paper refers to "overlay AS map" by comparision to
> "router-level map", not in relationship to "actual AS map" (which is
> not mentioned, to my knowledge).
>
> Thus the "overlay" part refers to overlaying ASes onto routers and
> links.
They actually compare their "overlay AS map" to "the actual AS map" obtained from BGP routing table dump at approximately the same period of time.
In fact, the figure 1 in http://citeseer.nj.nec.com/tangmunarunkit01internet.html paper state that both the maps have qualitatively similar macroscopic properties.
Unfortunately, the authors have failed to explain or provide references to what is "actual AS map"
-Ramesh
From akulkarn@usc.edu Tue Feb 4 21:16:20 2003
From: akulkarn@usc.edu (anupam kulkarni)
Date: Tue, 04 Feb 2003 13:16:20 -0800
Subject: [Csci551-talk] questions regarding Gao's paper
Message-ID: <129446b12961d5.12961d5129446b@usc.edu>
i have a couple of questions from Gao's paper.
1. i do not understand the significance of "Assumption P" could anyoneexplain it to me and
could guildline B be just as effective without that assumption.
2. the authors say that the local pref of a route that has a backup link needs to be the
lowest among all possible links. they further say that it is not necessary for the backup
link to be the first hop. I cannot figure out how a BGP speaker can find out about a backup
link if it in not the first hop. can anyone suggest me some example of backup links for an
AS that are not next hops for some destination.
From rmgandhi@usc.edu Wed Feb 5 22:47:42 2003
From: rmgandhi@usc.edu (ravi gandhi)
Date: Wed, 05 Feb 2003 14:47:42 -0800
Subject: [Csci551-talk] Reply to Anupam's questions (Gao's Paper)
Message-ID: <17d026e17d54ff.17d54ff17d026e@usc.edu>
1.Assumption P is essential to keep the system stable. In the example that the author provides, if the assumption is not kept, then by Guideline B, (1,0) is preferred by AS3 and there would be a loop and the system would not reach a stable state. By assuming P, the AS3 no longer prefers peer path (1,0) and the loop is avoided.
2.AS far as the 2nd doubt goes, there is no question of an AS trying to find out the length of the backup link. This is because i think a backup link is established by prior agreement between two ASes. SO that knowledge is provided beforehand.
Ravi M. Gandhi
MS(C.S.), University of Southern California
710, West 27th Street,
Apartment No.4,
Los Angeles,
CA 90007
Phone: 1-213-749-8983
webpage: http://www-scf.usc.edu/~rmgandhi/
From sarangar@usc.edu Thu Feb 6 00:12:52 2003
From: sarangar@usc.edu (ramesh sarangarajan)
Date: Wed, 05 Feb 2003 16:12:52 -0800
Subject: [Csci551-talk] questions regarding Gao's paper
Message-ID: <184be541849494.1849494184be54@usc.edu>
> 1. i do not understand the significance of "Assumption P" could
> anyoneexplain it to me and
> could guildline B be just as effective without that assumption.
Figure 5 & the description that follows answers this Q.
Guideline B relaxes the requirement that client routes should always have
higher local-pref than peer/provider route so that peer routes can have
local-pref as high that of client route but no more. Assumption P means
that the "AS cluster" graph is "acyclic".
( Note: AS cluster is a group of ASes that peer with each other )
> 2. the authors say that the local pref of a route that has a
> backup link needs to be the
> lowest among all possible links. they further say that it is not
> necessary for the backup
> link to be the first hop. I cannot figure out how a BGP speaker
> can find out about a backup
> link if it in not the first hop. can anyone suggest me some
> example of backup links for an
> AS that are not next hops for some destination.
>
The authors say that the back-up link does not have to be the "first hop"
[ in the AS path such as ( C, D, B, A ) ], but this doesn't mean that
the back-up link is not "single-hop".
Any link is a single-hop :)
-Ramesh
From sarangar@usc.edu Thu Feb 6 00:41:04 2003
From: sarangar@usc.edu (ramesh sarangarajan)
Date: Wed, 05 Feb 2003 16:41:04 -0800
Subject: [Csci551-talk] Gao-Rexford's BGP guidelines for multi-AS ISPs?
Message-ID: <1876661186f90e.186f90e1876661@usc.edu>
How do the BGP guidelines translate for ISPs with multiple ASes
( Ex: AT&T has 7018, 6341, 5074,....& possibly many others too ) ?
Would these ASes be peers? If so, isn't the "guideline B" being restrictive
( because peer routes can't have higher local-pref than client routes )
without offering economic incentives?
-Ramesh
From johnh@ISI.EDU Thu Feb 6 04:39:38 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Wed, 05 Feb 2003 20:39:38 -0800
Subject: [Csci551-talk] [Q] Internet Path Inflation Due...
In-Reply-To:
Message-ID: <200302060439.h164ddWT015422@dash.isi.edu>
On Mon, 03 Feb 2003 16:55:34 PST, ramesh sarangarajan wrote:
>> >I cannot quite get the difference between the "actual AS map" and
>> the AS overlay map.
>>
>> I believe the paper refers to "overlay AS map" by comparision to
>> "router-level map", not in relationship to "actual AS map" (which is
>> not mentioned, to my knowledge).
>>
>> Thus the "overlay" part refers to overlaying ASes onto routers and
>> links.
>
>They actually compare their "overlay AS map" to "the actual AS map" obtained
I chatted with Ramesh Govindan, one of the co-authors of the paper,
and he confirmed the distinction that I made quoted above about AS map
vs. router-level mpa.
-John Heidemann
From johnh@ISI.EDU Thu Feb 6 04:46:58 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Wed, 05 Feb 2003 20:46:58 -0800
Subject: [Csci551-talk] mailing list message formats
Message-ID: <200302060446.h164kw8I015477@dash.isi.edu>
I would like to encourage students to use plain text messages when
they post to the class mailing list, or both plain text and html text.
What format your mailers sends is usually an option.
Unfortunately, it seems like some recent recent e-mail programs (I
think some version of MS-Outlook) send mail as html-only. It's worth
figuring out how to turn this "feature" off (or at least to html+plain
text) since many people on the Internet prefer plain text (for
example, it's often easier to reply to), and some people can't easily
read html-only messages.
(All this goes ten-fold for sending MS-Word documents as the only
contents e-mail, but that's another story. :-)
-John Heidemann
From gnawali@usc.edu Thu Feb 6 07:27:10 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Wed, 05 Feb 2003 23:27:10 -0800
Subject: [Csci551-talk] HW1 Q3
Message-ID: <200302060727.h167RAlf022466@enl.usc.edu>
Some of you are curious about how to test your code on a little-endian
machine. For this all you need to do is compile your code in that
platform (Intel etc) and run it.
I have updated my FAQ page with this information.
From sbhavsar@usc.edu Sun Feb 9 00:18:57 2003
From: sbhavsar@usc.edu (siddharth bhavsar)
Date: Sat, 08 Feb 2003 16:18:57 -0800
Subject: [Csci551-talk] HW1 Q3
Message-ID: <22951e22afe7.22afe722951e@usc.edu>
apart from correctness are there any other criteria on which(eg code structure , efficiency etc) the
question will be graded .
Thanks
Siddharth
----- Original Message -----
From: Omprakash Gnawali
Date: Wednesday, February 5, 2003 11:27 pm
Subject: [Csci551-talk] HW1 Q3
>
> Some of you are curious about how to test your code on a little-endian
> machine. For this all you need to do is compile your code in that
> platform (Intel etc) and run it.
>
> I have updated my FAQ page with this information.
>
>
From gnawali@usc.edu Sun Feb 9 01:26:29 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Sat, 08 Feb 2003 17:26:29 -0800
Subject: [Csci551-talk] HW1 Q3
In-Reply-To: Your message of Sat, 08 Feb 2003 16:18:57 -0800
Message-ID: <200302090126.h191QTDA026285@enl.usc.edu>
1. Grading of HW1 Q3 will be based on correctness and correctness only
provided you are using the appropriate htons etc functions to do
conversion.
2. Please do not send emails to csci551-talk@isi.edu and to
gnawali@usc.edu at the same time. If you are sending an email to the
list, just send it to csci551-talk@isi.edu otherwise I get two emails
and I get confused.
you wrote:
> apart from correctness are there any other criteria on which(eg code structure , ef
*ficiency etc) the
> question will be graded .
> Thanks
> Siddharth
> ----- Original Message -----
> From: Omprakash Gnawali
> Date: Wednesday, February 5, 2003 11:27 pm
> Subject: [Csci551-talk] HW1 Q3
>
> >
> > Some of you are curious about how to test your code on a little-endian
> > machine. For this all you need to do is compile your code in that
> > platform (Intel etc) and run it.
> >
> > I have updated my FAQ page with this information.
> >
> >
>
From sbhavsar@usc.edu Mon Feb 10 01:30:05 2003
From: sbhavsar@usc.edu (siddharth bhavsar)
Date: Sun, 09 Feb 2003 17:30:05 -0800
Subject: [Csci551-talk] HW1 Q3
Message-ID: <3cf61b3d2aad.3d2aad3cf61b@usc.edu>
Here's a good animation of how a new startup is making money on BGP shortcomings .
http://www.routescience.com/technology/tec_fla_dem.html
Siddharth Bhavsar
----- Original Message -----
From: Omprakash Gnawali
Date: Saturday, February 8, 2003 5:26 pm
Subject: Re: [Csci551-talk] HW1 Q3
>
> 1. Grading of HW1 Q3 will be based on correctness and correctness only
> provided you are using the appropriate htons etc functions to do
> conversion.
>
> 2. Please do not send emails to csci551-talk@isi.edu and to
> gnawali@usc.edu at the same time. If you are sending an email to the
> list, just send it to csci551-talk@isi.edu otherwise I get two emails
> and I get confused.
>
> you wrote:
> > apart from correctness are there any other criteria on which(eg
> code structure , ef
> *ficiency etc) the
> > question will be graded .
> > Thanks
> > Siddharth
> > ----- Original Message -----
> > From: Omprakash Gnawali
> > Date: Wednesday, February 5, 2003 11:27 pm
> > Subject: [Csci551-talk] HW1 Q3
> >
> > >
> > > Some of you are curious about how to test your code on a
> little-endian
> > > machine. For this all you need to do is compile your code in that
> > > platform (Intel etc) and run it.
> > >
> > > I have updated my FAQ page with this information.
> > >
> > >
> >
>
>
From gnawali@usc.edu Tue Feb 11 03:55:30 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Mon, 10 Feb 2003 19:55:30 -0800
Subject: [Csci551-talk] Re: assignment word limit 551
Message-ID: <200302110355.h1B3tU4U028216@enl.usc.edu>
There is no word/sentence limit for the assignment unless stated
explicitly. You should take as many words and setences as necessary to
fully answer the questions. I am looking for content rather than
quantity. If there is enough content, even bullet-point answers will
get full credit.
> i wish to inquire as to what is the word limit on 1,2 questions on the assignment o
*r atleast what is the expected length.
>
From gnawali@usc.edu Tue Feb 11 03:57:28 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Mon, 10 Feb 2003 19:57:28 -0800
Subject: [Csci551-talk] HW1 Q3
Message-ID: <200302110357.h1B3vSSW028240@enl.usc.edu>
Your question is answered on the TA's FAQ page which you can access
from the course site. Towards the bottom of the course home page, you
will see a link called "TA web page".
> Where is the FAQ page? I still am unsure abotu how to check the code on little endi
*an.
> Do we need to detect the kind of system?
From johnh@ISI.EDU Wed Feb 12 05:09:30 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Tue, 11 Feb 2003 21:09:30 -0800
Subject: [Csci551-talk] questions regarding Gao's paper
In-Reply-To: <184be541849494.1849494184be54@usc.edu>
Message-ID: <200302120509.h1C59UtM025828@dash.isi.edu>
On Wed, 05 Feb 2003 16:12:52 PST, ramesh sarangarajan wrote:
>> 1. i do not understand the significance of "Assumption P" could
>> anyoneexplain it to me and
>> could guildline B be just as effective without that assumption.
>
>Figure 5 & the description that follows answers this Q.
>Guideline B relaxes the requirement that client routes should always have
>higher local-pref than peer/provider route so that peer routes can have
>local-pref as high that of client route but no more. Assumption P means
>that the "AS cluster" graph is "acyclic".
>( Note: AS cluster is a group of ASes that peer with each other )
Yes, Assumption P is required to guarantee that there are no cycles.
>
>> 2. the authors say that the local pref of a route that has a
>> backup link needs to be the
>> lowest among all possible links. they further say that it is not
>> necessary for the backup
>> link to be the first hop. I cannot figure out how a BGP speaker
>> can find out about a backup
>> link if it in not the first hop. can anyone suggest me some
>> example of backup links for an
>> AS that are not next hops for some destination.
Presumably the BGP speaker knows there is a backup link either through
out-of-band means, or through a community set. This is one of the BGP
attributes we didn't talk about. The distinction about single or
multi-hop backup links is not important for class.
-John Heidemann
From johnh@ISI.EDU Wed Feb 12 05:13:32 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Tue, 11 Feb 2003 21:13:32 -0800
Subject: [Csci551-talk] Gao-Rexford's BGP guidelines for multi-AS ISPs?
In-Reply-To: <1876661186f90e.186f90e1876661@usc.edu>
Message-ID: <200302120513.h1C5DWUY025876@dash.isi.edu>
On Wed, 05 Feb 2003 16:41:04 PST, ramesh sarangarajan wrote:
>How do the BGP guidelines translate for ISPs with multiple ASes
>( Ex: AT&T has 7018, 6341, 5074,....& possibly many others too ) ?
>
>Would these ASes be peers? If so, isn't the "guideline B" being restrictive
>( because peer routes can't have higher local-pref than client routes )
>without offering economic incentives?
They could treat other ASes that they happen to own however they like.
Note that section 6.2 (complex AS relationships) indicates that you
can apply rules per-prefix (not per-AS) and so they have considerable
flexibility.
-John Heidemann
From johnh@ISI.EDU Wed Feb 12 05:22:38 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Tue, 11 Feb 2003 21:22:38 -0800
Subject: routescience [was Re: [Csci551-talk] HW1 Q3 ]
In-Reply-To: <3cf61b3d2aad.3d2aad3cf61b@usc.edu>
Message-ID: <200302120522.h1C5McF5025944@dash.isi.edu>
On Sun, 09 Feb 2003 17:30:05 PST, siddharth bhavsar wrote:
>Here's a good animation of how a new startup is making money on BGP shortcomings .
>
>http://www.routescience.com/technology/tec_fla_dem.html
>
>Siddharth Bhavsar
Interesting commercial.
Something to ask yourself, though, is what if everyone was changing
their routing every minute based on observed performance?
-John Heidemann
From johnh@ISI.EDU Wed Feb 12 05:29:57 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Tue, 11 Feb 2003 21:29:57 -0800
Subject: [Csci551-talk] Re: assignment word limit 551
In-Reply-To: <200302110355.h1B3tU4U028216@enl.usc.edu>
Message-ID: <200302120529.h1C5TvGN025971@dash.isi.edu>
On Mon, 10 Feb 2003 19:55:30 PST, Omprakash Gnawali wrote:
>
>There is no word/sentence limit for the assignment unless stated
>explicitly. You should take as many words and setences as necessary to
>fully answer the questions. I am looking for content rather than
>quantity. If there is enough content, even bullet-point answers will
>get full credit.
>
>
>> i wish to inquire as to what is the word limit on 1,2 questions on the assignment o
> *r atleast what is the expected length.
>>
As Om says, if you can answer briefly and correctly (but covering all
the points raised in the question), even with bullets, you can get
full credit.
I would go further to suggest that if you're writing a whole lot, you
should ask yourself if you're really answering the question well. If
you find you have a book-long answer, you've probably missed the point
of the question. (And if it's clear that you've missed the point of
the question you can expect to be graded accordingly.)
-John Heidemann
From sarangar@usc.edu Wed Feb 12 13:30:05 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Wed, 12 Feb 2003 05:30:05 -0800
Subject: routescience [was Re: [Csci551-talk] HW1 Q3 ]
In-Reply-To: <200302120522.h1C5McF5025944@dash.isi.edu>
References: <3cf61b3d2aad.3d2aad3cf61b@usc.edu>
<200302120522.h1C5McF5025944@dash.isi.edu>
Message-ID: <20030212053005.0000610a.sarangar@usc.edu>
On Tue, 11 Feb 2003 21:22:38 -0800
"John Heidemann" wrote:
> On Sun, 09 Feb 2003 17:30:05 PST, siddharth bhavsar wrote:
> >Here's a good animation of how a new startup is making money on BGP
> >shortcomings .
> >
> >http://www.routescience.com/technology/tec_fla_dem.html
> >
> >Siddharth Bhavsar
>
> Interesting commercial.
>
> Something to ask yourself, though, is what if everyone was changing
> their routing every minute based on observed performance?
>
> -John Heidemann
>
"Chaos" - not convergence.
Application-level-overlays like "Akamai" would be a far better solution (but only for content delivery) than messing around with the network routes.
-Ramesh
From johnh@ISI.EDU Wed Feb 12 15:43:15 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Wed, 12 Feb 2003 07:43:15 -0800
Subject: routescience [was Re: [Csci551-talk] HW1 Q3 ]
In-Reply-To: <20030212053005.0000610a.sarangar@usc.edu>
Message-ID: <200302121543.h1CFhF6d028687@dash.isi.edu>
On Wed, 12 Feb 2003 05:30:05 PST, Ramesh Sarangarajan wrote:
>On Tue, 11 Feb 2003 21:22:38 -0800
>"John Heidemann" wrote:
>
>> On Sun, 09 Feb 2003 17:30:05 PST, siddharth bhavsar wrote:
>> >Here's a good animation of how a new startup is making money on BGP
>> >shortcomings .
>> >
>> >http://www.routescience.com/technology/tec_fla_dem.html
>> >
>> >Siddharth Bhavsar
>>
>> Interesting commercial.
>>
>> Something to ask yourself, though, is what if everyone was changing
>> their routing every minute based on observed performance?
>>
>> -John Heidemann
>>
>
>"Chaos" - not convergence.
>
>Application-level-overlays like "Akamai" would be a far better solution (but only for content delivery) than messing around with the network routes.
How does doing load-senstive routing at the app level
make it more (or less) stable than at the network level?
-John Heidemann
From sarangar@usc.edu Wed Feb 12 16:31:38 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Wed, 12 Feb 2003 08:31:38 -0800
Subject: routescience [was Re: [Csci551-talk] HW1 Q3 ]
In-Reply-To: <200302121543.h1CFhF6d028687@dash.isi.edu>
References: <20030212053005.0000610a.sarangar@usc.edu>
<200302121543.h1CFhF6d028687@dash.isi.edu>
Message-ID: <20030212083138.00006784.sarangar@usc.edu>
On Wed, 12 Feb 2003 07:43:15 -0800
"John Heidemann" wrote:
> On Wed, 12 Feb 2003 05:30:05 PST, Ramesh Sarangarajan wrote:
> >On Tue, 11 Feb 2003 21:22:38 -0800
> >"John Heidemann" wrote:
> >
> >> On Sun, 09 Feb 2003 17:30:05 PST, siddharth bhavsar wrote:
> >> >Here's a good animation of how a new startup is making money on BGP
> >> >shortcomings .
> >> >
> >> >http://www.routescience.com/technology/tec_fla_dem.html
> >> >
> >> >Siddharth Bhavsar
> >>
> >> Interesting commercial.
> >>
> >> Something to ask yourself, though, is what if everyone was changing
> >> their routing every minute based on observed performance?
> >>
> >> -John Heidemann
> >>
> >
> >"Chaos" - not convergence.
> >
> >Application-level-overlays like "Akamai" would be a far better solution (but only for content delivery) than messing around with the network routes.
>
> How does doing load-senstive routing at the app level
> make it more (or less) stable than at the network level?
>
> -John Heidemann
>
As app level routing don't modify the routing tables, it can be deployed easily (without having to be agreed by everybody running the networks).
Besides, if something goes wrong, we can always reconfigure our app level overlay or (if needed) revert to the original source of the content by
bypassing the overlay. Proxy caches have been around for a long time & if the caches get polluted, we can always get content the from original source.
But the major limitation of content overlays is that they just do only that & don't generalize to other apps. As the web & file-sharing are currently
the most common apps, this limitation is not serious atleast now.
-Ramesh
From gnawali@usc.edu Wed Feb 12 17:20:42 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Wed, 12 Feb 2003 09:20:42 -0800
Subject: routescience [was Re: [Csci551-talk] HW1 Q3 ]
In-Reply-To: Your message of Wed, 12 Feb 2003 08:31:38 -0800
Message-ID: <200302121720.h1CHKg2E015031@enl.usc.edu>
Related work: reading number 71 for our class. "Resilient Overlay
Networks" argues that it is possible to build a generic routing
overlay (without co-operation from unwilling ISPs). The catch is, you
better have some friends around the world living in different Internet
clouds. Also, think about what happens if everyone starts using RON.
> > >
> > >Application-level-overlays like "Akamai" would be a far better solution (but onl
*y for content delivery) than messing around with the network routes.
> >
> > How does doing load-senstive routing at the app level
> > make it more (or less) stable than at the network level?
> >
> > -John Heidemann
> >
>
> As app level routing don't modify the routing tables, it can be deployed easily (wi
*thout having to be agreed by everybody running the networks).
>
> Besides, if something goes wrong, we can always reconfigure our app level overlay o
*r (if needed) revert to the original source of the content by
> bypassing the overlay. Proxy caches have been around for a long time & if the cache
*s get polluted, we can always get content the from original source.
>
> But the major limitation of content overlays is that they just do only that & don't
* generalize to other apps. As the web & file-sharing are currently
> the most common apps, this limitation is not serious atleast now.
>
> -Ramesh
From teli@usc.edu Wed Feb 12 20:32:27 2003
From: teli@usc.edu (parthik teli)
Date: Wed, 12 Feb 2003 12:32:27 -0800
Subject: [Csci551-talk] Q for HW1
Message-ID: <88421887e1b6.87e1b6884218@usc.edu>
i have some question for HW1.
-- can i write the paper is "idea and analysis".. because it has introduce a idea and analyze it with different values....
-- can somethig about Q3...
my prog can un ob aludra... when i run it on VC i need to change some header files. so i want to ask uu that... do we need to make our program on both plateform.. like.. first it will detect the system.. and if VC-- Aludra.. and accordingly it will include file. ?? my basic code can run on both machine... just i am confused about this.. right now. i change my header file manually... do i need to make it something like automated ?
-Parthik Teli
======================
Hard work is the way to "Zenith".
Faith is the 'Base' of "Relation"
From mordani@usc.edu Wed Feb 12 22:12:58 2003
From: mordani@usc.edu (rohit mordani)
Date: Wed, 12 Feb 2003 14:12:58 -0800
Subject: [Csci551-talk] HW1 Q3 question
Message-ID: <9155d59126de.9126de9155d5@usc.edu>
HI,
HW1 Q3 gives us a hint to use memcpy. Do I have to use memcpy? My program works without using memcpy. If I dont use memcpy will I lose some points?
Thanks
Rohit
From gnawali@usc.edu Thu Feb 13 07:39:11 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Wed, 12 Feb 2003 23:39:11 -0800
Subject: [Csci551-talk] HW1 Q3 question
In-Reply-To: Your message of Wed, 12 Feb 2003 14:12:58 -0800
Message-ID: <200302130739.h1D7dBgc005157@enl.usc.edu>
Please use memcpy. If you don't use it you will not encounter some of
the issues that we wanted you guys to think about.
you wrote:
> HI,
> HW1 Q3 gives us a hint to use memcpy. Do I have to use memcpy? My program works without using memcpy.
* If I dont use memcpy will I lose some points?
> Thanks
> Rohit
>
>
From gnawali@usc.edu Thu Feb 13 07:44:47 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Wed, 12 Feb 2003 23:44:47 -0800
Subject: [Csci551-talk] Q for HW1
In-Reply-To: Your message of Wed, 12 Feb 2003 12:32:27 -0800
Message-ID: <200302130744.h1D7ilHr005219@enl.usc.edu>
I think the question would be trivial if you guys were allowed to say
it is "A and B and C and D" if A, B, C, and D are the only
possibilities. This is an approach I suggest. Think about the types of
the paper, and think about which category is the best match for the
paper and say the paper is of this type because it has ...
We are aware that there can be multiple answers for some papers and
the classification is not necessarily a disjoint partition.
As I mention in my FAQ page, you should make sure your program
compiles on aludra. I will test your program in an Intel Linux
platform and will make changes to the header myself if I need to. If
your program does not compile on aludra, you lose lots of points.
you wrote:
> i have some question for HW1.
> -- can i write the paper is "idea and analysis".. because it has introduce a idea and analyze it with
* different values....
> -- can somethig about Q3...
> my prog can un ob aludra... when i run it on VC i need to change some header files. so i want to as
*k uu that... do we need to make our program on both plateform.. like.. first it will detect the system.
*. and if VC-- Aludra.. and accordingly it will include file. ?? my basic code can run on both machine..
*. just i am confused about this.. right now. i change my header file manually... do i need to make it s
*omething like automated ?
> -Parthik Teli
> ======================
> Hard work is the way to "Zenith".
> Faith is the 'Base' of "Relation"
>
From shetye@usc.edu Thu Feb 13 20:21:31 2003
From: shetye@usc.edu (Siddharth Shetye)
Date: Thu, 13 Feb 2003 12:21:31 -0800
Subject: [Csci551-talk] testing hw1 program on Linux quickly
References: <200302122005.h1CK53D17893@gamma.isi.edu>
Message-ID: <001201c2d39d$80c7a910$06107e18@darkstar>
Hi,
This isin't about doubts in any paper, but just a small tip on getting linux
up and running really quick. I found this very useful to check my code for
HW1 if it ran on Linux over Intel ... took 4 minutes to boot and 1 minute to
confirm that it did. Thought maybe for this or future assignments people
could use this without much headache.
So try www.knoppix.com
Its a Linux distribution based on Debian which runs completely off a single
CD ! i.e. Boot thru the CD and you boot directly into KDE 3.0 with several
apps installed and functional.
Plus Points
- No need to (re)partition your hard disk
- No configuration files to mess with
- Excellent hardware detection mechanism (my roommates USB based cable
modem, a uncommon piece of hardware, was detected and installed! A shared
Cable network was autodetected and was up and running without any
intervention)
- gcc pre-installed
- OpenOffice pre-installed (equivalent of MS Office)
- quick to work with
Minuses
- Can't make it a "permanent"/regular Linux OS on your hard disk (without
much hacking, that is)
- hence non-persistent
- Limited App support (get just whats on the CD)
I was quite impressed by this linux flavour ... hence this email !
Siddharth
From sarangar@usc.edu Thu Feb 13 23:28:55 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Thu, 13 Feb 2003 15:28:55 -0800
Subject: [Csci551-talk] Knoppix Linux [was: testing hw1 program on Linux quickly]
In-Reply-To: <001201c2d39d$80c7a910$06107e18@darkstar>
References: <200302122005.h1CK53D17893@gamma.isi.edu>
Message-ID: <5.1.0.14.2.20030213151657.02c37e68@email.usc.edu>
Yes, the hardware support of Knoppix (live cd linux) is truely amazing....&
it brings up the KDE desktop without asking a single question...the
commercial distros must be ashamed of themselves...
My $0.02:
If you have a wireless access point, set the network SSID to linux-wlan
(the default for wlan-ng package in knoppix), enable DHCP & disable WEP.
You can start browsing in Konqueror without any config in knoppix. (In fact
I used knoppix for installing gentoo - the default kernel in gentoo didn't
support my network cards & I use linksys wpc-11 card)
But, as Siddarth has said, knoppix is no replacement for linux installed on
the hard drive. It's at best a "Swiss army knife" linux.
-Ramesh
At 12:21 PM 2/13/2003 -0800, you wrote:
>Hi,
>
>This isin't about doubts in any paper, but just a small tip on getting linux
>up and running really quick. I found this very useful to check my code for
>HW1 if it ran on Linux over Intel ... took 4 minutes to boot and 1 minute to
>confirm that it did. Thought maybe for this or future assignments people
>could use this without much headache.
>
>So try www.knoppix.com
>Its a Linux distribution based on Debian which runs completely off a single
>CD ! i.e. Boot thru the CD and you boot directly into KDE 3.0 with several
>apps installed and functional.
>
>Plus Points
>- No need to (re)partition your hard disk
>- No configuration files to mess with
>- Excellent hardware detection mechanism (my roommates USB based cable
>modem, a uncommon piece of hardware, was detected and installed! A shared
>Cable network was autodetected and was up and running without any
>intervention)
>- gcc pre-installed
>- OpenOffice pre-installed (equivalent of MS Office)
>- quick to work with
>
>Minuses
>- Can't make it a "permanent"/regular Linux OS on your hard disk (without
>much hacking, that is)
>- hence non-persistent
>- Limited App support (get just whats on the CD)
>
>I was quite impressed by this linux flavour ... hence this email !
>
>Siddharth
From johnh@ISI.EDU Thu Feb 13 23:33:13 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Thu, 13 Feb 2003 15:33:13 -0800
Subject: [Csci551-talk] Q for HW1
In-Reply-To: <200302130744.h1D7ilHr005219@enl.usc.edu>
Message-ID: <200302132333.h1DNXE1J011311@dash.isi.edu>
On Wed, 12 Feb 2003 23:44:47 PST, Omprakash Gnawali wrote:
>
>I think the question would be trivial if you guys were allowed to say
>it is "A and B and C and D" if A, B, C, and D are the only
>possibilities. This is an approach I suggest. Think about the types of
>the paper, and think about which category is the best match for the
>paper and say the paper is of this type because it has ...
I would add, please make sure that you justify your choice in part c.
Just saying "it is A" is not a complete answer.
>We are aware that there can be multiple answers for some papers and
>the classification is not necessarily a disjoint partition.
>
>As I mention in my FAQ page, you should make sure your program
>compiles on aludra. I will test your program in an Intel Linux
>platform and will make changes to the header myself if I need to. If
>your program does not compile on aludra, you lose lots of points.
People use #ifdef's to make code portable to multiple platforms. All
C compliers define certain pre-processor symbols on different
platforms to allow you to do things specific to that platform. You
can use ifdefs to disable Windows-specific functionality on
Solaris/Linux.
-John Heidemann
From johnh@ISI.EDU Thu Feb 13 23:36:00 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Thu, 13 Feb 2003 15:36:00 -0800
Subject: [Csci551-talk] HW1 Q3 question
In-Reply-To: <200302130739.h1D7dBgc005157@enl.usc.edu>
Message-ID: <200302132336.h1DNa0t8011346@dash.isi.edu>
On Wed, 12 Feb 2003 23:39:11 PST, Omprakash Gnawali wrote:
>
>Please use memcpy. If you don't use it you will not encounter some of
>the issues that we wanted you guys to think about.
However, you are not *required* to use memcpy. If you don't use it,
you will not lose points because of that.
But you *are* required to use ntohs. You should expect to lose points
for not using that function. (Even if your program otherwise works.)
Om is right in that if memcpy and ntohs work well together, so you're
encouraged to use memcpy.
-John Heidemann
>
>you wrote:
>> HI,
>> HW1 Q3 gives us a hint to use memcpy. Do I have to use memcpy? My program works without using memcpy.
> * If I dont use memcpy will I lose some points?
>> Thanks
>> Rohit
>>
>>
From sarangar@usc.edu Fri Feb 14 20:26:15 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Fri, 14 Feb 2003 12:26:15 -0800
Subject: [Csci551-talk] TCP SACK in Linux
Message-ID: <5.1.0.14.2.20030214115931.00b15010@email.usc.edu>
Linux kernel 2.4 has TCP SACK, which can be enabled as simple as writing
"1" to /proc/sys/net/ipv4/tcp_sack
The other interesting performance tuning options are mentioned in
http://www.psc.edu/networking/perf_tune.html#Linux
-Ramesh
From sarangar@usc.edu Fri Feb 14 21:18:37 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Fri, 14 Feb 2003 13:18:37 -0800
Subject: [Csci551-talk] "pathchar" docs?
Message-ID: <5.1.0.14.2.20030214130941.00baba00@email.usc.edu>
Is there any good documentation (not the usage notes) about how the
"pathchar" tool works?
I remember using that a couple of years ago (in my college in India - by
then we had a very low (by today's standards) bandwidth link (64 Kbps) to
the Internet which was always heavily congested & the tool still gave
amazingly accurate estimation of the link bandwidths - at least I knew the
bandwidth of the next few links of the ISP).
If a tool estimated bandwidth when the link is not being utilized, it's
perceivable; but how does "pathchar" estimate bandwidth even in a heavily
congested link?
Unfortunately, I could not find docs explaining how "pathchar" works. If
someone knows about such docs, please tell me.
-Ramesh
From shshanks@usc.edu Sat Feb 15 21:57:19 2003
From: shshanks@usc.edu (shshank sharma)
Date: Sat, 15 Feb 2003 13:57:19 -0800
Subject: [Csci551-talk] Re: "pathchar" docs?
Message-ID: <2eaf322ee00f.2ee00f2eaf32@usc.edu>
> 2. "pathchar" docs? (Ramesh Sarangarajan)
> Is there any good documentation (not the usage notes) about how
> the "pathchar" tool works?
Pathchar basically estimates the throughput and latency based on repeated
RTT measurements trying to reach the minimum possible RTT.
You might find this acrobat file useful.
http://delivery.acm.org/10.1145/310000/301582/p222-downey.pdf?key1=301582&key2=6225435401&coll=portal&dl=ACM&CFID=11111111&CFTOKEN=2222222
Even msri-talk.pdf available at ftp://ftp.ee.lbl.gov/pathchar/ talks about this.
Let me know if the links dont work for you, I can send the files across.
- Sharma.
From sarangar@usc.edu Sun Feb 16 05:01:21 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Sat, 15 Feb 2003 21:01:21 -0800
Subject: [Csci551-talk] bandwidth measurement by non-privileged users [Was: [Csci551-talk]
Re: "pathchar" docs?]
In-Reply-To: <2eaf322ee00f.2ee00f2eaf32@usc.edu>
Message-ID: <5.1.0.14.2.20030215204732.00b15508@email.usc.edu>
Thanks for the info.
But, I'm uncomfortable with pathchar because it's available only as a
binary. When I searched for similar tools, I found the following:
- pipechar http://www-didc.lbl.gov/pipechar/
- nettimer http://mosquitonet.stanford.edu/~laik/projects/nettimer/
These tools seem to be actively maintained and available in source form.
But, both of these require admin privileges to run. The pipechar docs say
that there can't be multiple instances running at the same time (meaning
they can't be accessed programmatically).
Is there some tool (or library) , which can be accessed programmatically by
non-privileged (l)users?
BTW, even a rough estimate of the available end-to-end bandwidth would be
enough for my needs.
-Ramesh
At 01:57 PM 2/15/2003 -0800, you wrote:
> > 2. "pathchar" docs? (Ramesh Sarangarajan)
> > Is there any good documentation (not the usage notes) about how
> > the "pathchar" tool works?
>
>Pathchar basically estimates the throughput and latency based on repeated
>RTT measurements trying to reach the minimum possible RTT.
>You might find this acrobat file useful.
>
>http://delivery.acm.org/10.1145/310000/301582/p222-downey.pdf?key1=301582&key2=6225435401&coll=portal&dl=ACM&CFID=11111111&CFTOKEN=2222222
>
>
>Even msri-talk.pdf available at ftp://ftp.ee.lbl.gov/pathchar/ talks about
>this.
>Let me know if the links dont work for you, I can send the files across.
>
>- Sharma.
From johnh@ISI.EDU Mon Feb 17 18:04:19 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Mon, 17 Feb 2003 10:04:19 -0800
Subject: [Csci551-talk] TCP SACK in Linux
In-Reply-To: <5.1.0.14.2.20030214115931.00b15010@email.usc.edu>
Message-ID: <200302171804.h1HI4JDJ021782@dash.isi.edu>
On Fri, 14 Feb 2003 12:26:15 PST, Ramesh Sarangarajan wrote:
>Linux kernel 2.4 has TCP SACK, which can be enabled as simple as writing
>"1" to /proc/sys/net/ipv4/tcp_sack
Cool. Apparently I've been running SACK for a while without knowing
it. :-)
-John Heidemann
From johnh@ISI.EDU Mon Feb 17 18:12:04 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Mon, 17 Feb 2003 10:12:04 -0800
Subject: [Csci551-talk] "pathchar" docs?
In-Reply-To: <5.1.0.14.2.20030214130941.00baba00@email.usc.edu>
Message-ID: <200302171812.h1HIC4Vb021916@dash.isi.edu>
On Fri, 14 Feb 2003 13:18:37 PST, Ramesh Sarangarajan wrote:
>Is there any good documentation (not the usage notes) about how the
>"pathchar" tool works?
There are multiple good papers about pathchar.
A good place to search is INSPEC, a commercial technical paper
database that USC subscribes to.
At USC, you can get to INSPEC from
and then look under "Quicklinks".
-John Heidemann
From shshanks@usc.edu Tue Feb 18 05:05:11 2003
From: shshanks@usc.edu (Shshank)
Date: Mon, 17 Feb 2003 21:05:11 -0800
Subject: [Csci551-talk] RE: bandwidth measurement by non-privileged users
In-Reply-To: <200302162005.h1GK5DD15235@gamma.isi.edu>
Message-ID:
But, I'm uncomfortable with pathchar because it's available only as a
binary. When I searched for similar tools, I found the following:
- pipechar http://www-didc.lbl.gov/pipechar/
- nettimer http://mosquitonet.stanford.edu/~laik/projects/nettimer/
These tools seem to be actively maintained and available in source form.
But, both of these require admin privileges to run. The pipechar docs say
that there can't be multiple instances running at the same time (meaning
they can't be accessed programmatically).
> I'm not sure I got what you mean by 'accessing programmatically'. Do you
mean access to the source code, or tool metrics being available to
non-privileged users ? Please do explain that a bit more.
Is there some tool (or library) , which can be accessed programmatically by
non-privileged (l)users?
BTW, even a rough estimate of the available end-to-end bandwidth would be
enough for my needs.
> If you havent already, you might want to check out iperf, for throughput
measurements, lost packet counts, and more !
Source code and lots of binaries are available here:
http://dast.nlanr.net/Projects/Iperf/
- Sharma
From sarangar@usc.edu Tue Feb 18 18:33:32 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Tue, 18 Feb 2003 10:33:32 -0800
Subject: [Csci551-talk] RE: bandwidth measurement by non-privileged users
In-Reply-To:
References: <200302162005.h1GK5DD15235@gamma.isi.edu>
Message-ID: <5.1.0.14.2.20030218102324.00b15650@email.usc.edu>
>>The pipechar docs say
>>that there can't be multiple instances running at the same time (meaning
>>they can't be accessed programmatically).
> I'm not sure I got what you mean by 'accessing programmatically'. Do you
> mean access to the source code, or tool metrics being available to
> non-privileged users ? Please do explain that a bit more.
The following is an excerpt from pipechar web site:
A Key Issue in using pipechar is that you should NEVER run multiple
pipechar on a host. Running multiple pipechar from a single host
simultaneously will cause them hang each other. Pipechar is not designed to
do such job and is not designed for programming and development.
This is network layer (L3) v.s. transport layer (L4) issue. The
multiplexing capability is at L4 only. Because all switch and routers are
either link layer (L2) or network layer or both devices, and the control
message is via ICMP in L3, there is no way to let multiple pipechar
programs running on a host to share and De-multiplexing one income ICMP
stream. An incoming in ICMP packet will be sent to all open RAW sockets.
Question: traceroute is also a network layer tool, how it works if I run
multiple traceroute simultaneously from a host?
Answer: traceroute does not time the income ICMP, it only needs to know
what types of income ICMP packets are. In this case, traceroute may spend
time to toss those ICMP packets not belong to itself. PxCHAR is different,
if we spend a lot of time to filter out irrelevant ICMP packets, timing is
screwed up. That is it. Happy :-)
HTH
-Ramesh
From sarangar@usc.edu Tue Feb 18 18:40:35 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Tue, 18 Feb 2003 10:40:35 -0800
Subject: [Csci551-talk] bandwidth measurement by non-privileged users
[Was: [Csci551-talk] Re: "pathchar" docs?]
In-Reply-To: <5.1.0.14.2.20030215204732.00b15508@email.usc.edu>
References: <2eaf322ee00f.2ee00f2eaf32@usc.edu>
Message-ID: <5.1.0.14.2.20030218103534.00b25dd0@email.usc.edu>
>Is there some tool (or library) , which can be accessed programmatically
>by non-privileged (l)users?
>
>BTW, even a rough estimate of the available end-to-end bandwidth would be
>enough for my needs.
At last, I found a good (entirely app-level) tool for estimating E2E
bottleneck bandwidth.
It's explained in the following paper & talk:
http://www.acm.org/sigcomm/sigcomm2002/papers/e2ebw.pdf
http://detache.cmcl.cs.cmu.edu/broadcast_archive/sigcomm2002/ManishJain_250k.asx
-Ramesh
From teli@usc.edu Wed Feb 19 22:54:00 2003
From: teli@usc.edu (parthik teli)
Date: Wed, 19 Feb 2003 14:54:00 -0800
Subject: [Csci551-talk] SACK Problem
Message-ID: <659e766e14.66e14659e7@usc.edu>
I cannot understand why in SACK, upon receiveng Partial ACK.. Pipe is decremented by 2... can somebody please explain it ?
-Parthik
======================
Hard work is the way to "Zenith".
Faith is the 'Base' of "Relation"
From johnh@ISI.EDU Thu Feb 20 04:40:31 2003
From: johnh@ISI.EDU (John Heidemann)
Date: Wed, 19 Feb 2003 20:40:31 -0800
Subject: [Csci551-talk] Project A
Message-ID: <200302200440.h1K4eV1U015028@dash.isi.edu>
Project A has been posted to the class web site.
-John Heidemann
From gnawali@usc.edu Fri Feb 21 07:00:10 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Thu, 20 Feb 2003 23:00:10 -0800
Subject: [Csci551-talk] hw1 grade
Message-ID: <200302210700.h1L70AfO013248@enl.usc.edu>
I just sent out grades for hw1. Let me know if you didn't receive it.
- om_p
From gnawali@usc.edu Fri Feb 21 17:31:10 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Fri, 21 Feb 2003 09:31:10 -0800
Subject: [Csci551-talk] hw1 q3 test cases
Message-ID: <200302211731.h1LHVAkR018883@enl.usc.edu>
The four test cases used for grading HW1Q3 are now on my website.
From gnawali@usc.edu Sun Feb 23 05:05:40 2003
From: gnawali@usc.edu (Omprakash Gnawali)
Date: Sat, 22 Feb 2003 21:05:40 -0800
Subject: [Csci551-talk] Timer API
Message-ID: <200302230505.h1N55eRF005415@enl.usc.edu>
Makefile in Timer API fixed. The fix is related to getting C
application to compile in Linux platform. Consequently, if you want to
use this code in C, you will have to link your program the way the
Makefile does. For C++, there were no changes.
From sarangar@usc.edu Fri Feb 28 21:06:52 2003
From: sarangar@usc.edu (Ramesh Sarangarajan)
Date: Fri, 28 Feb 2003 13:06:52 -0800
Subject: [Csci551-talk] a controversial paper: TCP circuits
Message-ID: <5.1.0.14.2.20030228124554.00baaff0@email.usc.edu>
I'm sure this paper -
http://citeseer.nj.nec.com/molinero-fernandez02tcp.html - would have
ignited a lot of controversy. This makes a case for integrating circuit
switching with packet switching.
I think they make a pretty good case for switching circuits for every TCP flow:
1. Bandwidth is no longer the "precious" entity - it is the buffers in routers.
2. Moore's law for processors is moving much slower than the rate of
doubling of bandwidth (every 7 months as opposed to every 18 months for
processors).
3. Already we are over-provisioning bandwidth & the original argument for
packet switching (achieving max utilization by statistical multiplexing) is
less relevant currently.
4. Already the Core of the Internet (as well as the last-mile) is carried
on circuits (SONET, etc)
4. The robustness (routing around failures) in Internet is lesser (seconds
to minutes) currently because of complex routing protocols which take a
long time to converge - whereas SONET recovers in less than 50 ms.
5. They argue that the state & signalling for TCP switching is manageable
with current hardware tech.
Very interesting indeed - if their proposal becomes successful, all the
previous efforts in QoS provisioning in packet-switched Internet would
become less relevant.
But, as I'm not really an expert, I suspect there could be something wrong
in their idea.
Any ideas on why this radically different way of doing thing won't actually
work?
-Ramesh
PS: A slightly different version of the same paper was presented in
Hot-Interconnects 2001 conference:
http://tiny-tera.stanford.edu/~nickm/papers/TCPSwitch-HotI_final_paper.pdf