You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Mark <st...@gmail.com> on 2010/08/13 18:32:32 UTC
TimeUUID vs Epoch
I'm a little confused on when I should be using TimeUUID vs Epoch/Long
when I want columns ordered by time. I know it sounds strange and the
obvious choice should be TimeUUID but I'm not sure why that would be
preferred over just using the Epoch stamp?
The pretty much seem to accomplish the same thing however when using
Epoch I can easily use a start/end range for querying across times. Can
this be accomplished using TimeUUID?
Would someone also explain how TimeUUID is actually sorted? Im confused
on how its actually compared. Thanks!
Re: TimeUUID vs Epoch
Posted by Mark <st...@gmail.com>.
On 8/13/10 10:01 AM, Sylvain Lebresne wrote:
> True, it's more annoying to do because you need to construct a UUID whose time
> part is what you need and the rest of the bytes are zeros (excepted for the
> version bit that should be one for the UUID to be valid if I remember
> correctly).
> And usually, UUID generator don't give you that, you'll have to 'do it
> yourself',
> which is annoying (but not very hard if you look a UUID version 1 layout).
>
> --
> Sylvain
>
> On Fri, Aug 13, 2010 at 6:54 PM, Mark<st...@gmail.com> wrote:
>
>> On 8/13/10 9:38 AM, Sylvain Lebresne wrote:
>>
>>> As long as time sorting is involved, you'll the same ordering if you
>>> use Epoch/Long
>>> or TimeUUID. The difference is between the ties. If when you insert
>>> two values at
>>> the exact same time, you want to have only one stay, then you want
>>> LongType.
>>> If however you don't want to merge such inserts, then you want TimeUUID
>>> since
>>> they assure you that even if generated at the exact same time, the key
>>> will be different
>>> (plus if your UUID generator is not too bad, it will give you better
>>> precision than the
>>> epoch/long one).
>>>
>>> --
>>> Sylvain
>>>
>>> On Fri, Aug 13, 2010 at 6:32 PM, Mark<st...@gmail.com> wrote:
>>>
>>>
>>>> I'm a little confused on when I should be using TimeUUID vs Epoch/Long
>>>> when
>>>> I want columns ordered by time. I know it sounds strange and the obvious
>>>> choice should be TimeUUID but I'm not sure why that would be preferred
>>>> over
>>>> just using the Epoch stamp?
>>>>
>>>> The pretty much seem to accomplish the same thing however when using
>>>> Epoch I
>>>> can easily use a start/end range for querying across times. Can this be
>>>> accomplished using TimeUUID?
>>>>
>>>> Would someone also explain how TimeUUID is actually sorted? Im confused
>>>> on
>>>> how its actually compared. Thanks!
>>>>
>>>>
>>>>
>>>>
>>>>
>> So long story short you can give a start/end range when using TimeUUID?
>>
>> For example I am storing a bunch of records keyed by the current date
>> "20100813". Each column is a TimeUUID. If I wanted to get all the columns
>> that between some arbitrary time.. say 6am - 9am I can get that?
>>
>> Using Long I can just use a start of "1281704400000000" and a finish of
>> "1281715200000000" but Im unsure of how to represent that as a TimeUUID.
>>
>>
Thanks for the help. Im using UUID version 1 so Ill have to look into
how this would be accomplished. Im using the SimpleUUID ruby gem so if
anyone knows how this is done please chime in :)
Re: TimeUUID vs Epoch
Posted by Sylvain Lebresne <sy...@yakaz.com>.
True, it's more annoying to do because you need to construct a UUID whose time
part is what you need and the rest of the bytes are zeros (excepted for the
version bit that should be one for the UUID to be valid if I remember
correctly).
And usually, UUID generator don't give you that, you'll have to 'do it
yourself',
which is annoying (but not very hard if you look a UUID version 1 layout).
--
Sylvain
On Fri, Aug 13, 2010 at 6:54 PM, Mark <st...@gmail.com> wrote:
> On 8/13/10 9:38 AM, Sylvain Lebresne wrote:
>>
>> As long as time sorting is involved, you'll the same ordering if you
>> use Epoch/Long
>> or TimeUUID. The difference is between the ties. If when you insert
>> two values at
>> the exact same time, you want to have only one stay, then you want
>> LongType.
>> If however you don't want to merge such inserts, then you want TimeUUID
>> since
>> they assure you that even if generated at the exact same time, the key
>> will be different
>> (plus if your UUID generator is not too bad, it will give you better
>> precision than the
>> epoch/long one).
>>
>> --
>> Sylvain
>>
>> On Fri, Aug 13, 2010 at 6:32 PM, Mark<st...@gmail.com> wrote:
>>
>>>
>>> I'm a little confused on when I should be using TimeUUID vs Epoch/Long
>>> when
>>> I want columns ordered by time. I know it sounds strange and the obvious
>>> choice should be TimeUUID but I'm not sure why that would be preferred
>>> over
>>> just using the Epoch stamp?
>>>
>>> The pretty much seem to accomplish the same thing however when using
>>> Epoch I
>>> can easily use a start/end range for querying across times. Can this be
>>> accomplished using TimeUUID?
>>>
>>> Would someone also explain how TimeUUID is actually sorted? Im confused
>>> on
>>> how its actually compared. Thanks!
>>>
>>>
>>>
>>>
>
> So long story short you can give a start/end range when using TimeUUID?
>
> For example I am storing a bunch of records keyed by the current date
> "20100813". Each column is a TimeUUID. If I wanted to get all the columns
> that between some arbitrary time.. say 6am - 9am I can get that?
>
> Using Long I can just use a start of "1281704400000000" and a finish of
> "1281715200000000" but Im unsure of how to represent that as a TimeUUID.
>
Re: TimeUUID vs Epoch
Posted by Mark <st...@gmail.com>.
On 8/13/10 9:38 AM, Sylvain Lebresne wrote:
> As long as time sorting is involved, you'll the same ordering if you
> use Epoch/Long
> or TimeUUID. The difference is between the ties. If when you insert
> two values at
> the exact same time, you want to have only one stay, then you want LongType.
> If however you don't want to merge such inserts, then you want TimeUUID since
> they assure you that even if generated at the exact same time, the key
> will be different
> (plus if your UUID generator is not too bad, it will give you better
> precision than the
> epoch/long one).
>
> --
> Sylvain
>
> On Fri, Aug 13, 2010 at 6:32 PM, Mark<st...@gmail.com> wrote:
>
>> I'm a little confused on when I should be using TimeUUID vs Epoch/Long when
>> I want columns ordered by time. I know it sounds strange and the obvious
>> choice should be TimeUUID but I'm not sure why that would be preferred over
>> just using the Epoch stamp?
>>
>> The pretty much seem to accomplish the same thing however when using Epoch I
>> can easily use a start/end range for querying across times. Can this be
>> accomplished using TimeUUID?
>>
>> Would someone also explain how TimeUUID is actually sorted? Im confused on
>> how its actually compared. Thanks!
>>
>>
>>
>>
So long story short you can give a start/end range when using TimeUUID?
For example I am storing a bunch of records keyed by the current date
"20100813". Each column is a TimeUUID. If I wanted to get all the
columns that between some arbitrary time.. say 6am - 9am I can get that?
Using Long I can just use a start of "1281704400000000" and a finish of
"1281715200000000" but Im unsure of how to represent that as a TimeUUID.
Re: TimeUUID vs Epoch
Posted by Sylvain Lebresne <sy...@yakaz.com>.
As long as time sorting is involved, you'll the same ordering if you
use Epoch/Long
or TimeUUID. The difference is between the ties. If when you insert
two values at
the exact same time, you want to have only one stay, then you want LongType.
If however you don't want to merge such inserts, then you want TimeUUID since
they assure you that even if generated at the exact same time, the key
will be different
(plus if your UUID generator is not too bad, it will give you better
precision than the
epoch/long one).
--
Sylvain
On Fri, Aug 13, 2010 at 6:32 PM, Mark <st...@gmail.com> wrote:
> I'm a little confused on when I should be using TimeUUID vs Epoch/Long when
> I want columns ordered by time. I know it sounds strange and the obvious
> choice should be TimeUUID but I'm not sure why that would be preferred over
> just using the Epoch stamp?
>
> The pretty much seem to accomplish the same thing however when using Epoch I
> can easily use a start/end range for querying across times. Can this be
> accomplished using TimeUUID?
>
> Would someone also explain how TimeUUID is actually sorted? Im confused on
> how its actually compared. Thanks!
>
>
>