RE: Ants Data Server
- From: "Laimutis Nedzinskas" <Laimutis.Nedzinskas@xxxxxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Tue, 19 Sep 2006 16:05:12 -0000
There are no miracles in this world.
A few quotes from
http://www.primebase.com/xt/download/pbxt_white_paper.pdf :
PBXT is transactional, but only writes data once namely, to the log.
This has a number of consequences:
* Firstly the obvious: writing less, saves times.
* The database table data actually resides in the logs (how we get to
the table data is explained below in the section "Getting a Handle on
the Problem").
* Since table data is not stored in any particular table file, the size
of a table is unlimited.
* Since log files are written sequentially, PBXT only ever writes
transaction data sequentially.
* Since data is only written sequentially, variable sized rows pose no
problem to PBXT. In fact, there is actually no need to specify a maximum
size for column types such as VARCHAR, VARBIN or BLOBs.
* Since log file data is written immediately, records in the cache are
never "dirty" (i.e. must be flushed before being freed). This greatly
speeds up and simplifies cache management.
* Since log files are never updated, PBXT never modifies table data in
place. This means that instead of updating a record, PBXT writes a new
record.
* Since database table data is never updated rollback never has to undo
a change.
* Since data is never overwritten, the internal data structure is
inherently more stable than a standard implementation.
PBXT takes this idea a step further. In a standard implementation one
transaction log is used. To reduce contention in a high concurrency
environment, PBXT uses one log per thread of execution. This effectively
means each transaction has its own log.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Remigiusz Sokolowski
Sent: 19. september 2006 15:50
To: Roger.Xu@xxxxxxxxxxx
Cc: Oracle-L@Freelists. Org (E-mail)
Subject: Re: Ants Data Server
Roger Xu wrote:
> They claim there is rarely write-locks rows and all indexes, queues
> and other internal datastructures are fully concurrent, with no lock
> waiting ever!
> How is this possible?
>
> -----Original Message-----
> *From:* oracle-l-bounce@xxxxxxxxxxxxx
> [mailto:oracle-l-bounce@xxxxxxxxxxxxx]*On Behalf Of *Roger Xu
> *Sent:* Monday, September 18, 2006 10:05 AM
> *To:* Oracle-L@Freelists. Org (E-mail)
> *Subject:* OT: Ants Data Server
>
> Hi List,
>
> Anyone has experience with Ants Data Server?
>
> http://www.ants.com
>
> Roger Xu
>
well, I suppose it can be done in similar way as it is in case of
http://www.primebase.com/xt/ under link above You'll find engine for
mysql, which author claims to run without locks, while being
transactional
regards
Remigiusz
--
---------------------------------------
Remigiusz Sokolowski <rems@xxxxxxxx>
WP/PTI/DIP/ZAB (+04858) 52 15 770
---------------------------------------
--
//www.freelists.org/webpage/oracle-l
Fyrirvari/Disclaimer
http://www.landsbanki.is/disclaimer
--
//www.freelists.org/webpage/oracle-l
Other related posts:
