Bugs item #1593305, was opened at 2006-11-09 12:05
Message generated for change (Settings changed) made by boncz
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=482468&aid=1593305&group_id=56967

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: PF/runtime
Group: Pathfinder CVS Head
>Status: Closed
>Resolution: Wont Fix
Priority: 3
Private: No
Submitted By: Peter Boncz (boncz)
Assigned to: Sjoerd Mullender (sjoerd)
Summary: PF: perf issue: map_pid logging

Initial Comment:
map_pid is currently logged in full for each transaction.

in the 10GB xmark, we map-pid has size 8000, which will
times 1+8+8 bytes, thus 136KB per single update.

given that map_pid is [void,oid], it would be better
not log the head, and use the array-write functionality
in the logger (that is faster and shorter than a log
call for each bun).

[ SIDE ISSUE: it would also be faster and shorter to
use array-write for all other [oid,any] delta bats;
splitting them by column ]

additionally, given that the mapi_pid contains page
numbers, we could convert it in te log to an [void,int]
bat, or maybe even a [void,sht] bat, and dynamically
cast upwards such datatypes during log recovery. In the
10GB case, this would reduce map_pid logging overhead
to to 16KB per query.

[ SIDE ISSUE AGAIN: array write save the log mode byte
plus this upcasting will probably save  additional
space (on oids) on all other delta bats as well.

An interesting idea would be to use the X100
compression interface for writing to the log,
especially when all deltas are written with the array
interface. ]

But, for map_pid, it would be better to encode inserts
and deletes, rather than log the entire thing:

e.g. delta [oid,oid] head = insertpos, oid newtail
where 
- insertpos is the original position after 
  which the insert is done. 
- multiple identical insertpos entries indicate multipl
 tuples inserted at that pos
- negative (higest bit set) insertpos are deletes

----------------------------------------------------------------------

>Comment By: Peter Boncz (boncz)
Date: 2009-04-07 10:56

Message:
will not be fixed anymore

----------------------------------------------------------------------

Comment By: Stefan Manegold (stmane)
Date: 2007-06-04 20:38

Message:
Logged In: YES 
user_id=572415
Originator: NO

Peter, Sjoerd,

is this one still an open issues after all the recent XQUF code
changes/improvements?

Stefan


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=482468&aid=1593305&group_id=56967

------------------------------------------------------------------------------
This SF.net email is sponsored by:
High Quality Requirements in a Collaborative Environment.
Download a free trial of Rational Requirements Composer Now!
http://p.sf.net/sfu/www-ibm-com
_______________________________________________
Monetdb-bugs mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/monetdb-bugs

Reply via email to