- users-list - monetdb.org

[MonetDB-users] Re: MonetDB Questions
by Martin Kersten 17 May '05

17 May '05

Hello Mason, thank you for your interest in using MonetDB. Enclosed a quick response to your questions. If I miss something, someone from the monetdb-users list is likely to fill in the details. M Sharp wrote: > Hello Martin, > > We are interested in evaluating MonetDB for use with our product (a > potential customer of ours told us queries ran 5-6 times faster than on That;s what we know to be true for especially decision support applications. > PostgreSQL). Basically we would install MonetDB on several nodes, and > partition data amongst them, for data warehousing usage (more info about > our product can be found at http://www.extendb.com) > > I have looked through the documentation and website, but have some > questions I was hoping you could help answer (as the project lead). If > it is not too much trouble, I would greatly appreciate it if you could > take a few minutes to answer. > > 1. Size limitation. If I understand correctly, there is no size > limitation for tables or databases, just that a single column is limited > to 2 GB. Is that correct? Also, when do you believe this restriction > will be removed? The size limitation of 2 GB holds for 32-bit platforms, it is already removed for 64 bit machines. > > 1a. I see something in the config file about 3GB of virtual memory. If > the database (or table) is larger than that, does it negatively impact > performance? The performance degradation starts when MonetDB has to access BATs larger then the available memory, but won;t become slower than what you would expect in traditional database systems. The judicious use of the BATs mean that in practice of decision support you have a much lower chance hitting an IO bottleneck. > > 2. Multiple databases. I noticed in the config file that only one > database can be used at a time. Can we start mulitple MonetDB instances > to support multiple databases, just using a different port? There can be one Mserver per database directory, and ofcourse multiple Mserver instances per platform. > > 3. Backups. Is there any backup tool? Also, can you do incremental > backups? No, we are looking forward to people interesting in developing such supportive tools. > > 4. Is there point in time recovery? The SQL frontend ensures a log of all committed transactions to ensure stability. If you use the system at the MIL layer you have to schedule that work yourself. > > 5. Do you support savepoints (or subtranscations), not just > transactions? The SQL transaction semantics is based on the requirements in SQL99 > > 6. What kind of locking capability is offered; DB, table or row? Are > there different isolation levels? There are no separate isolation levels. The policy in MonetDB is to assume a cooperative environment, which means that (SQL) compilers have to explicity lock/release resources to support concurrent use from clients. Internally, the system is fully configured to safely run with tens of concurrent threads thereby protecting the integrity of the database structures (but not their content). > > 7. Is there a separate high-speed loader utility, or must INSERTs be > used? yes there is. > > 8. Are checkpoints used? And is it configurable? No particular support for taking a checkpoint > > 9. Are there utilities by which you can determine corruption? ... And > repair corruption? The additional tools have not been needed so far. The SQL log was sufficient to deal with the rare cases during development crashes. > > > > Thank you very much in advance for taking the time to answer these > questions. Also, please feel free to provide any other thoughts you have > regarding using MonetDB and handling large databases (~ 1 TB). It is the sizes we are after, but not on 32-bit platforms. > > Regards, > > Mason Sharp > ExtenDB >

1 0

[MonetDB-users] VOIDs and multiple BATs
by Edmund Dengler 09 May '05

09 May '05

Greetings! Just doing some more digging around MonetDB, and have a couple of questions. (1) If I am managing multiple BATs that are (supposedly) synced, how do I manage appends and the syncing? For example, I have a_p1 = BAT[void, int] and a_p2 = BAT[void, int] (a_p1 = <a> type things with property p1) (a_p2 = <a> type things with property p2) I wish to insert a new <a>=(1,2) thing. Do I do? bat("a_p1").append(1); bat("a_p2").append(2); How do I ensure that the "voids" are always synced? Is this just "assumed"? What if I fail in the middle of operations (such that a_p1 has an extra row) and stuck with dirty tables? Do I do a global "abort();" or a "clean();" to get back to a safe state? Will this reset the seqbase? (Is this documented someplace? I am trying to look at the architecture docs but cannot find anything that concretely answers similar questions.) (2) How does sync() work? Do I do a: sync(bat("a_p1"), bat("a_p2")); once and only once? After every update/append to the tables? There is the phrase "When two BATs effectively contain the same sequence of head elements, we call them 'synced'. This is implemented by storing a very large OID for each column. An update to the column destroys this OID.". This sounds as if any update/append breaks the correspondence, and also that it will limit the number of rows in the table (as an OID is a 32-bit number of 32-bit systems). If I have 10 properties I am tracking, would I pick one as the "primary" BAT, and sync() all others to it? Regards! Ed

3 5

[MonetDB-users] Counting open BATs
by Edmund Dengler 05 May '05

05 May '05

Greetings! In the architectural overview, I find the text: "Most memory resource problems in MonetDB are easily avoidable by minimizing your open BAT references (as discussed above) and making your access patterns cache friendly (see the radix module)." Is there anyplace that actually describes when a BAT is considered open? Is it a scope thing (BATs are closed as soon as they are out of scope)? Is a BAT open when I append/update? Does it close when I perform a commit? If not, is there a way to specify "I am done, clean it out"? I am noticing a large amount of open memory (approximately 500MB out of a 1GB system) being used by MonetDB. I have hand partitioned my BATs into groups using a naming convention, but I am wondering if during inserts I am somehow keeping open references to the BATs which may cause issues. Is there a way to find out where memory is being allocated? Which BATs are open? Also, I am setting some of the larger BATs to STORE_MMAP. I notice in the documentation that if I compress the heaps, then these must be set to STORE_MEM. Is there a reason that when uncompressing the heap, that the uncompressed version cannot be STORE_MMAP to avoid memory problems? Regards! Ed

2 3

[MonetDB-users] Use of C/C++ prepare statements
by Edmund Dengler 21 Apr '05

21 Apr '05

Greetings! Does anybody have any working example code using mapi_prepare/map_prepare_array/mapi_execute_array showing the use of a prepare along with array substitution? I can get a sample piece working using mapi_execute. Trying mapi_execute_array causes the test program to exit with an error code of 139 (the function does not seem to return, but causes an immediate exit). Using mapi_prepare/mapi_execute_array, the program runs, but nothing is changing in the database. Regards! Ed

2 1

[MonetDB-users] Questions
by Edmund Dengler 21 Apr '05

21 Apr '05

Howdy all! I had posted some questions regarding MonetDB, but have not received feedback on any of them. Is there a more appropriate list I should be posting to? Regards! Ed

3 2

[MonetDB-users] Crashing server
by Edmund Dengler 18 Apr '05

18 Apr '05

Greetings! Having some problems with the MonetDB server crashing. - Start a server - Have a client perform a lot of operations (_lots_ of BATs creation, bun insertions, etc). - Every 10,000 or so operations, perform a <commit();> on the client. Total operations: approximately 7500 BATs, 17,000,000 BUNs. - Do an <ls();> on the server to confirm all tables are clean. - Now perform an <abort();> on the server with every table clean. MonetDB>abort(); #GDKmmap(6553600) fail => BBPtrim(enter) usage[mem=736705664,vm=2137835808] # !mallinfo.arena = 35753984 !mallinfo.ordblks = 4454 !mallinfo.smblks = 0 !mallinfo.hblkhd = 122863616 !mallinfo.hblks = 33 !mallinfo.usmblks = 0 !mallinfo.fsmblks = 0 !mallinfo.uordblks = 15990816 !mallinfo.fordblks = 19763168 #GDKmmap(6553600) fail => BBPtrim(ready) usage[mem=736705664,vm=2137835808] # !mallinfo.arena = 35753984 ... (and so on) Doing a <ulimit -a> shows: core file size (blocks, -c) 0 data seg size (kbytes, -d) unlimited file size (blocks, -f) unlimited max locked memory (kbytes, -l) 32 max memory size (kbytes, -m) unlimited open files (-n) 1024 pipe size (512 bytes, -p) 8 stack size (kbytes, -s) 8192 cpu time (seconds, -t) unlimited max user processes (-u) 14336 virtual memory (kbytes, -v) unlimited Anybody have any ideas on why the server died? If I do it with less (_much_ less) operations, it at least succeeds (even though there is a lot of I/O occurring). Regards! Ed

2 2

[MonetDB-users] Question about OIDs and scaling
by Christopher Petrilli 31 Mar '05

31 Mar '05

I have been reading about MonetDB for a few days now, and have played around with it some and it seems to solve some of the ultra complex queries that I'd like to do in a time that is amazing. Congrats! One thing, though, that confuses me is some scaling issues. The data set that I'm looking at is potentially in the billions of traditional rows, with about 50 columns, so from my understanding that'd be 50 different BATs each with a billion or more items. My concern is that OIDs seem to be restricted to 2^32-1, although from looking at virtual OIDs, it's unclear whether in my situation that would be useful? Any thoughts on stuffing that much data into the system, especially how it might impact dealing with VM issues? I'm more than happy to say "this is the wrong approach" or system for this problem, as this is obviously a non-normal case. Chris -- | Christopher Petrilli | petrilli(a)gmail.com

2 4

[MonetDB-users] Bulk load to MonetDB
by Juan José van der Linden 24 Mar '05

24 Mar '05

Hi, I've compiled MonetDB and the SQL front-end on my PowerMac 2x2 G5. I've used fink to install the libraries needed. I would like to perform the query test that are available from http://benchw.sourceforge.net . How can I use the acsii_io module to bulk load the data files from the benchw project? Thanks in advance. Regards, Juan José.

2 1

[MonetDB-users] Compilation issue on OS X Panther
by Fabian 20 Mar '05

20 Mar '05

(I am not the one with the problem here, but reporting on behalf to stick to the protocol. Please ignore your user agent's default reply to address and send replies to the list, not to me.) The following build error appears for a user that compiles on a Mac: gcc -g -O2 -std=c99 -Wall -W -Werror-implicit-function-declaration -Werror -Wno-format -Wno-unused-function -Wno-unused-label -Wno-uninitialized -D_REENTRANT -o .libs/MapiClient MapiClient.o ReadlineTools.o -L/sw/lib ./.libs/libMapi.dylib /Users/rvjansen/Desktop/MonetDB/src/common/.libs/libstream.dylib -lreadline ../../../common/.libs/libmutils.dylib ../../../common/.libs/libstream.dylib /Users/rvjansen/Desktop/MonetDB/src/common/.libs/libmutils.dylib -lz /sw/lib/libiconv.dylib ld: Undefined symbols: _BZ2_bzclose _BZ2_bzflush _BZ2_bzopen _BZ2_bzread _BZ2_bzwrite make[7]: *** [MapiClient] Error 1 make[6]: *** [all] Error 2 make[5]: *** [all-recursive] Error 1 make[4]: *** [all-recursive] Error 1 make[3]: *** [all] Error 2 make[2]: *** [all-recursive] Error 1 make[1]: *** [all-recursive] Error 1 make: *** [all] Error 2 The user's tools versions: automake 1.9.3 aclocal 1.9.3 autoconf 2.59 libtool ? (there is no --version: default Apple distribution one)

3 9

[MonetDB-users] [Fwd: Re: cvs lock]
by Fabian 20 Mar '05

20 Mar '05

Here find attached the config.status file of the OSX compilation issue. about the mailing list settings stuff (which should be off the list, IMHO): there is a weird distinction between what developers think is useful and what users think is useful. For the sourceforge/Mailman case it seems to me that the former one is considered instead of the latter, which is -- if you expect users to use it -- a wrong assumption. Since SF is made for developers (you have to have a developers account for instance), SF is clearly NOT the place for user interaction. Proof of this theorem can be found in many projects that host a forum like structure on their main website, that often allows guest posting, which is clearly geared towards user usability. (how big do you want to have the barrier that prevents someone from reporting a problem today?)

1 0