Distributed Computing Q&A regarding the homework


More recent questions will be at the beginning (top) of the page.

Q&A regarding the material in class:






On Thu, 29 Apr 2004, Work Site wrote:

> Yehuda shalom,

>

> the meaning is that v and u are not being colored seperately and

> independently!

> they are colored together, and then when i move to a neighbor w of u, i have

> to

> consider the color of u, which has already been determined.Is it true ?

>

NO.

When you color v, you do not think of any other node. It is a

completely independent process.  Only when you want

to prove that the coloring is legal you show that the algorithm must have

chosen different colors for v and u (by observing the ring from the

previous email).  Notice that v can have many neighbors.

To prove that the coloring of v does not collide with any of its neighbors

colors you use many different rings with different sequences of ids.


Yehuda shalom,

>

> there is a something that i havn't understood yet :

> you wrote me :

> "Each time we color a node we take a new copy of a ring (simulating the

> ring on the side) assign the 2t+1 sequence of ids from the node label to

> nodes on the Ring and figure out what will be the color of the node in the

> middle of the 2t+1 sequence.  This color is then given to the

> corresponding node in G(2t+1)....."

>

> my question is :

> if there is an edge from v to u in G(2t+1), and i color now v, does the

> coloring of v have influence on the coloring of u ?

> if not, the meaning is that the nodes are colored independently, so why

> can't two nodes be colored the same ( without any connection to the prove of

> the coloring ) ?

The label of v in G(2t+1) is [x1, x2, ...., xt, xt+1, ..., x2t+1],

u neighbor of v label is        [x2, x3, ...., xt+1, xt+2, ..., x2t+2].

 

In any ring in which a sequence of nodes have the ids:

..., x1, x2, x3, ...., xt, xt+1, xt+2, ...., x2t+1, x2t+2, .... the

algorithm 6-colors node xt+1 and node xt+2 with different colors,

therefore the colors of v and u are different in G(2t+1).


First consider the answer to your second question and then the answer

to your first question.

 

2.  Coloring a node in G(2t+1):

The assumption is that there is a t rounds algorithm to 6-color nodes in a ring. 

After t rounds of such an algorithm on a

ring, each node on the ring was affected only by nodes that are at

distance t or less from it.  I.e., given the t nodes on each side of it, a node in

the ring can tell what would be its coloring in the Ring coloring

algorithm.  Now, we use this coloring to color the nodes of G(2t+1).

Each time we color a node we take a new copy of a ring (simulating the

ring on the side) assign the 2t+1 sequence of ids from the node label to

a continuous stretch of nodes on the Ring and figure out what will be

the color of the node in the middle of the 2t+1 sequence. 

This color is then given to the corresponding node in G(2t+1).

 

Now to your first question, why is this a legal coloring:

Because first observe that the sequence of ids in the label of one node in

G(2t+1) is a shift by one of the sequence of its neighboring node, with one

new id poping on the right.  Therefore, to prove that this is a legal coloring,

we take a ring and put the union of these two labels (i.e., a sequence of length

2t+2) on a ring and run the 6-coloring algorithm, then the color of

each of the two nodes in

G(2t+1) should be legal since the coloring in the ring is legal.

 

On Wed, 28 Apr 2004,  wrote:

 

 

> Yehuda shalom,

> i have two more questions:

>

> 1. u wrote me : "To fix the color of v...."

>     i understand that each node is colored independently, so how can i

> promise that 2 neighbours won't be colored by the same color ?

>

> 2. why do i need t nodes from each side(of x(t+1)), when i color a node v ?

>

> thanks,


> my SECOND question is, again :

>

> given an algorithm A for ring's 6-coloring in t iterations(a "black box"),

> how can i color G(2t+1) in 6 colors?

Recall that each node in G(2t+1) is labeled with a sequence of 2t+1

Ring-node-ids.  To fix the color of node v in G(2t+1) whose label is [x1,

x2, ..., x(t+1), ..., x(2t+1)] construct a ring with more than 2t+1 nodes,

and assign the ids x1, x2, ..., x(2t+1) to a continuous sequence of 2t+1 nodes

on the ring.  Now run A on the ring you have just constructed and color

node v with the color of node x(t+1) on the constructed ring.  Claim, this

is a 6 coloring of G(2t+1).  Prove this claim by contradiction: notice you

can construct a ring with a sequence [x1, x2, ..., x(2t+2)] of 2t+2 ids

on the ring such that the colors of v and its neighbor in G(2t+1) are the

colors of two neighbors on this ring.

 >

> thanks,


Notice there is a new solution on the web site for the Question 2 in HW 1. http://www.cs.tau.ac.il/~afek/solution1.103.html


> Yehuda shalom,

> and my questions are :

Answers below, and I hope to give again an overview in class next week.

> 1. in the last lesson, you actually proved that MIS cannot be maid in T

> iterations,

>     when T is constant, but you did not proved that it cannot be maid in

> less then log*n

>     iterations. is it true ?

We actually proved that there is no algorithm to color a ring (in

the model that we discussed, synchronous, unbounded message size) with 6

colors in less than log^* n (denoted log*n).  This is denoted o(log*n), i.e., there is no

algorithm that colors a ring with 6 colors in o(log*n) (little o notation). 

We did it by establishing two facts:

1.  that if there is such

an algorithm that color a ring with 6 colors in t<log*n then we can color

graph G(2t+1) (as defined in class) with 6 colors.  and

2.  that the chromatic number of G(h) greater equal log^h(n) (=log log

log ... log n,  taking log h times), and therefore in order for the chromatic number

of G(h) to be constant (like 6) h must be in the order of log*n (e.g.,

some constant times log*n, say 2 log*n).   Notice that this completes the

proof by the following argument:

By contradiction: Assume you have an algorithm A that colors

the ring with 6 colors in t' such that t'< log*n then I color G(t') with

6 colors a contradiction to point 2.  Notice we make no assumptions on A,

which makes the proof a strong one.  E.g., no assumption on message size.

> 2. i did not understand how you can use the MIS algorithm to color G(2t+1)

> in 6

>     colours.can u explain this again ?

>

> thanks,

In class I argued the other way around.  That is, that

given a 6 coloring of the ring then I can easily construct an MIS, as

follows:  Say the colors are c1, c2, c3, ..., c6.  Then I have 5 more

rounds, round-1, round-2,.... after we 6 color the ring:   In round-i,

i=1,...,5 all the nodes with color ci which are not yet marked OUT mark

themselves IN and tell all their immediate neighbors to mark themselves

OUT.  Claim (prove it to yourself) after the 5th round the IN marked nodes

are an MIS.