[9fans] [plan9mod] Query regarding venti scores

[9fans] [plan9mod] Query regarding venti scores

[anoop.anoop]

Hello All,

I have a query regarding the scores in venti. Here is my question:
Since the venti scores are SHA-1 hashes, can they be considered as
UUIDs? I have two different venti servers and am storing different
sets of databolcks on each. I am wondering whether a scenario will
arise where server A and server B gives the same score for completely
different datablocks.

Any help is greatly appreciated. Thanks in advance.

Regards,
Anoop

Re: [9fans] [plan9mod] Query regarding venti scores

[23hiro]

With today's amount of storage the probability of a collision is too
low for that to happen.
I read that in the Venti paper[1].
I would not consider them UUIDs though, since the same data blocks
will have the same sha-1 hash on both servers.
If that doesn't cause a problem you're fine

[1] http://plan9.bell-labs.com/sys/doc/venti/venti.pdf


On 2/6/09, anoop.anoop@gmail.com <anoop.anoop@gmail.com> wrote:
> Hello All,
>
> I have a query regarding the scores in venti. Here is my question:
> Since the venti scores are SHA-1 hashes, can they be considered as
> UUIDs? I have two different venti servers and am storing different
> sets of databolcks on each. I am wondering whether a scenario will
> arise where server A and server B gives the same score for completely
> different datablocks.
>
> Any help is greatly appreciated. Thanks in advance.
>
> Regards,
> Anoop
>
>

Re: [9fans] [plan9mod] Query regarding venti scores

[Anthony Sorace]

the score is the hash of the data in the block. the same block on two
servers will have the same score (important for replication).
different blocks have the same odds of having this same score (that
is, astronomically low) regardless of whether they're on one or
multiple servers.

that doesn't directly address their suitability as UUIDs. it seems
like an odd choice to me (certainly it's an expensive way of
generating them), although i guess it depends what your application
is.

Re: [9fans] [plan9mod] Query regarding venti scores

[erik quanstrom]

> that doesn't directly address their suitability as UUIDs. it seems
> like an odd choice to me (certainly it's an expensive way of
> generating them), although i guess it depends what your application
> is.

sha-1 is a standard uuid format.  since a uuid is 16 bytes and
sha-1 is 20, there should be plenty of room in there.  here's the
fully legal url:

http://en.wikipedia.org/wiki/UUID#Version_5_.28SHA-1_hash.29

- erik