Fw: [chord] replication in chord + dhash

[Frank Dabek]

----- Original Message ----- 
From: "Srinivas V." <srini at CS.Arizona.EDU>
To: "Frank Dabek" <fdabek at mit.edu>
Sent: Wednesday, May 28, 2003 1:20 AM
Subject: Re: [chord] replication in chord + dhash


> Hi
>
> Thanks for replying
>
> >     I'm assuming here that you are using the most recent CVS. In this
> > version the -n flag is unused.
>
> I'm using a CVS version that's a few weeks old.
>
> >     When half of the nodes in a system die, we would expect that some
blocks
> > will become unavailable, namely those blocks which had more than 7of
their
> > fragments stored on the failed nodes. The probability is a little messy
due
> > to the use of erasure codes, but you get the idea.
> >
> > How many blocks did you see unavailable?
>
> This seems consistent with what I observed. Less than half the blocks
> became unavailable. I ran a few more tests where only a small fraction of
> the dhash nodes were killed. No blocks were lost then.
>
>        srinivas
>
>