February 2005 - users-list

[MonetDB-users] questions regarding mil-sql interoperation and savepoint behavior
by Kinzer, Jason M (Jason) 26 Feb '05

26 Feb '05

Hi, I'm exploring Monet, and it's a very pleasant experience. I have a few questions however; please forgive the parade of ignorance that follows: - If one creates persistent void-head BATs via MIL, is there a way to them as columns of a SQL table/view. I noticed the parser had a rule allowing "CREATE VIEW ... AS BATS", but wasn't quite sure how to use it. - Conversely, if one has a set of tables, is there any easy way to find that mapping of columns to bat ids (e.g. so that one can lookup individual bats in MIL)? It appears this information is really only persisted in the log file, is that correct? If so, I assume the only other option would be to have the sql module loaded, and then access the temporary bats it creates that hold this metadata (I didn't notice that the module exported any MIL-accessible functions for grabbing the sql_column data)? - I am wondering how transaction chains are supposed to behave w.r.t. memory consumption. Specifically, I noticed that when executing "savepoint x" in the mapiclient, the heap size of the Monet process climbs almost monotonically upwards - jumping perhaps 350K per savepoint (every tx is completely empty) and hardly falling at all following releases or rollbacks (or even client disconnect for that matter). My understanding is that a savepoint is essentially a child transaction (thus copied from the parent), in which case I'm worried it may not be releasing all resources related to the tx and therefore leaking memory? Or is it simply that monet is allocating additional space from the OS, say for BATs related to the transaction, and this memory is becoming part of the internal pool and thus not being release back to the OS? I'd assume the second except that I thought it would reuse the space so that the heap size would stop climbing after a while, and I never did reach that poi! nt even after 100's of savepoints and an 30-fold increase in heap size. (BTW, I'm running the pre-build MonetDB 4.6.0/SQL 2.6.0 on Win2K). Many thanks for your time, and any responses. Cheers, Jason

1 0

[MonetDB-users] compile on Windows - pthread version
by Michael Schmid 24 Feb '05

24 Feb '05

Hello, i am a new user of MonetDB and want to share my problems and solutions with you. I try'ed to compile the source code version on Windows and run in a problem: the comiler give out an error: can not convert from pthread_t to int ...... After study some source code, and read the history of changes in the pthread distro, i read about changing this structure. I downloaded older versions of the Pthread lib, and finally: it compiles well. But only with pthreads-2004-06-22.exe or older, all newer version give the above error. From the Pthreads news "SNAPSHOT 2004-11-03 ------------------- DLLs produced from this snapshot cannot be used with older applications without recompiling the application, due to a change to pthread_t to provide unique POSIX thread IDs." i guess, this is the reason. You should make a hint in the HowToStart-Win32.txt to this, so that other users dont have the same problem. This problem occors on Win2000 and WinXP, with the the release and the CVS code. Do you have plans to change the code the new pthread_t structure, so that newer versions of the PThread distro can be used? Best regards, Michael Schmid

2 1

[MonetDB-users] Delete is slow?
by Edmund Dengler 14 Feb '05

14 Feb '05

Howdy! I'm taking some time to see what MonetDB is all about, and I have some questions re performance. I created 6 BATs (basically BAT[LNG, X] where X = INT, OID, TIMESTAMP, BLOB) to test. I loaded up about 250k rows (loading was quite reasonable, even though I was using the Perl interface to do it one row at a time). I am currently trying to delete the BUNs by doing a series of "bat("bat_name").delete();", and it is just taking forever (currently at over an hour!!). I/O is extremely minimal, CPU is nailed to 100% (this is running the latest release candidate off of sourceforge on a Linux 2.6 Gentoo system). Questions: Should delete be so slow? Is there a dependency between tables that is somehow forcing an N^2 or N^3 algorithm to run? Is there an equivalent to "truncate" other than destroy() and recreate? Regards! Ed

5 17

[MonetDB-users] ascii_io
by Edmund Dengler 11 Feb '05

11 Feb '05

Greetings! Looking at the ascii_io module, I can't seem to find a way for loading up multiple BATs where one column of the input is specified as the head value. Is there such an option, or is it always a VOID loadup? In particular, I have a << lng >> column as the primary key. So, using my current understanding, I would need to load up a set of dummy/temp BATs, and then using these load the data to the real BATs. Regards! Ed

3 5

[MonetDB-users] Design of bat layout
by Edmund Dengler 11 Feb '05

11 Feb '05

Greetings! I have a question concerning how to design the data structures for a particular problem. I have a set of records, each record may have 0 to 20 strings associated with it. If I was to use an SQL database, I would use a table such as: table strings ( long record_id, int index, string string ); I was thinking of using a BAT within a BAT, something like: BAT[ BAT[ lng, int ], str ] but this cannot be made persistent. Using: bat(lng, str).rename("string_1"); bat(lng, str).rename("string_2"); ... bat(lng, str).rename("string_20"); just feels wrong (if for no other reason, what if there is a 21st string in the future?). I could do: bat(lng::1, lng::2).rename("record_strings"); bat(lng::1, int).rename("record_string_index"); bat(lng::1, str).rename("record_string_string"); But this means I am essentially limiting my record IDs to not be the full rnage of a long. What would be the suggested structure to use in this case? Regards! Ed

1 0