create table #T
(
  I int,
  D datetime
)

create table #Result
(
  I int
)

insert into #T(I, D) values(1, '20121108 15:00'),
                           (1, '20121108 16:00'),
                           (2, '20121108 17:00'),
                           (1, '20121108 18:00'),
                           (3, '20121108 19:00'),
                           (3, '20121108 20:00'),
                           (2, '20121108 21:00')
declare @I     int,
        @I_Old int = 0

declare curD cursor for
  select I from #T order by D

open curD
while 1 = 1
begin
  fetch next from curD into @I
  if @@fetch_status != 0
    break

  if @I_Old != @I
  begin
    insert into #Result(I) values (@I)
    set @I_Old = @I
  end
end
close      curD
deallocate curD

select I from #Result

drop table #T
drop table #Result

select *
from

(select top 1 with ties I, D
from

(select *, 
row_number() over (order by D) - row_number() over (partition by I order by D) as rn
from @t ) a

order by row_number() over (partition by I, rn order by D)) b

order by D

I           D
----------- -----------------------
1           2012-11-08 15:00:00.000
2           2012-11-08 17:00:00.000
1           2012-11-08 18:00:00.000
3           2012-11-08 19:00:00.000
2           2012-11-08 21:00:00.000

--
-- Тестовый набор данных:
with
  t (i,d) as
    (
      select *
       from (values
               (1, '20121108 15:00'),
               (1, '20121108 16:00'),
               (1, '20121108 16:30'),
               (2, '20121108 17:00'),
               (2, '20121108 17:10'),
               (2, '20121108 17:20'),
               (2, '20121108 17:30'),
               (1, '20121108 18:00'),
               (1, '20121108 18:15'),
               (1, '20121108 18:25'),
               (3, '20121108 19:00'),
               (3, '20121108 20:00'),
               (2, '20121108 21:00'),
               (2, '20121108 21:05'),
               (2, '20121108 21:10') 
             ) as v(i,d)
             
      )
--
-- Основной запрос
select i, max(d) as beg_date, min(d) as end_date, count(1) as i_cnt
  from (
         select i,d, 
                sum(start_of_group) over(order by d) as grp_id -- Формирование идентификатора группы последовательно идущих данных
           from (
                  select t.*, 
                         case lag(i) over(order by d) 
                           when i
                             then 0
                           else 1 
                         end as start_of_group -- Флаг начала группы последовательно идущих данных
                    from t
                ) as v0
       ) as v1
 group by i, grp_id