Re: [boost] [guid] - vversion 1, 4 and in one class?

From: Andy (atompkins_at_[hidden])
Date: 2007-02-28 10:26:10

<Oliver.Kowalke_at_[hidden]> wrote in
news:B1EAFF2DAE7658488B631F25813CD91FD819DC_at_[hidden]: 

> Hello Andy, 
> 
>> My implementation of version 4 (random-number-based) guids 
>> may not be unique in space and time.  It is based on 
>> boost::mt19937 to generate random numbers and it is seeded 
>> with std::time(0).  Cryptographic-quality random numbers 
>> would reduce the probability of repeated values.
>> 
>> Using version 5 (name-based) guid can be used to ensure that 
>> guids are unique in space (guids generated on several 
>> computers at the same time) and time.  Generate a guid for 
>> each computer as the namespace_guid and pass a counter as the 
>> name to the create function.
> 
> Because creating uuid v5 seams too expensive for me I could let one
> special computer (in my project == master) generate guids (unique in
> space).
> So I've make a decision wether uuid v4 or v1 should be used in order
> to be uinque in time?!
> Or do you have another suggestion?

I have not done any profiling/timing on any of the functions that create 
a guid.  A note about multiple threads: all 3 versions are thread safe, 
but the random-number-based (version 4) and the time-based (version 1) 
function need a mutex to ensure that only one thread is creating a guid 
at a time.  The name-based (version 5) does not need a mutex.

As to decide between time-based (version 1) and random-number-based 
(version 4) for which would be unique in time, I would expect the time-
based (version 1) to be more unique in time since only the node is 
generated randomly and only the first time a guid is generated (until the 
program is restarted).  Whereas the random-number-based (version 4) is 
all just random data.

Andy.
> 
> Regards,Oliver
> _______________________________________________
> Unsubscribe & other changes:
> http://lists.boost.org/mailman/listinfo.cgi/boost 
>