Re: [sqlite] Index creation on huge table will never finish.
--- [EMAIL PROTECTED] wrote: > Joe Wilson <[EMAIL PROTECTED]> wrote: > > > > As for the stats from sqlite3_analyzer, they seem to be in the right > > ballpark. > > But I'm not sure its heuristic accounts for rows that are significantly > > larger > > than the page size, though. In such cases I am seeing higher than expected > > fragmentation after a VACUUM. This is just a guess, of course. > > I'm not sure sqlite3_analyzer does any of the fragmentation > measurement right. For that matter, how do you measure > fragmentation with a number? (Suggestions are welcomed.) > > I may yet just yank that whole fragmentation measurement > idea. With non-volatile RAM drives getting larger and cheaper by the day, you may not need to worry about fragmentation, fsync and disk-seek time in a year or two. Expecting? Get great news right away with email Auto-Check. Try the Yahoo! Mail Beta. http://advision.webevents.yahoo.com/mailbeta/newmail_tools.html - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Joe Wilson <[EMAIL PROTECTED]> wrote: > > As for the stats from sqlite3_analyzer, they seem to be in the right ballpark. > But I'm not sure its heuristic accounts for rows that are significantly larger > than the page size, though. In such cases I am seeing higher than expected > fragmentation after a VACUUM. This is just a guess, of course. > I'm not sure sqlite3_analyzer does any of the fragmentation measurement right. For that matter, how do you measure fragmentation with a number? (Suggestions are welcomed.) I may yet just yank that whole fragmentation measurement idea. -- D. Richard Hipp <[EMAIL PROTECTED]> - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
--- [EMAIL PROTECTED] wrote: > Joe Wilson <[EMAIL PROTECTED]> wrote: > > > > See also: Changes to support fragmentation analysis in sqlite3_analyzer. > > http://www.sqlite.org/cvstrac/chngview?cn=3634 > > > > I'm not real sure those patches are working right. > I need to revisit that whole fragmentation analysis > thing before the next release. Somebody please > remind me What do you suspect is not working right? I've run the new CVS VACUUM on large databases without any apparent ill effect and it's noticably faster than the previous VACUUM implementation. As for the stats from sqlite3_analyzer, they seem to be in the right ballpark. But I'm not sure its heuristic accounts for rows that are significantly larger than the page size, though. In such cases I am seeing higher than expected fragmentation after a VACUUM. This is just a guess, of course. Expecting? Get great news right away with email Auto-Check. Try the Yahoo! Mail Beta. http://advision.webevents.yahoo.com/mailbeta/newmail_tools.html - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Joe Wilson <[EMAIL PROTECTED]> wrote: > > See also: Changes to support fragmentation analysis in sqlite3_analyzer. > http://www.sqlite.org/cvstrac/chngview?cn=3634 > I'm not real sure those patches are working right. I need to revisit that whole fragmentation analysis thing before the next release. Somebody please remind me -- D. Richard Hipp <[EMAIL PROTECTED]> - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
P Kishor wrote: Mac/Unix person meself, but the Windows XP sort is pretty darn good as well. I'll take a look. Last time I used it it was useless. Win9x days? these days (especially for a one off) I'd probably go straight to doing it in Python to avoid x-platform syntax issues. Martin - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Ah yes, I should read more carefully :) Thanks, right, I was actually guaranteeing uniqueness originally by just fetching and then inserting only if there wasn't a match (I needed a rowid if the row existed anyway). Now I'm guaranteeing uniqueness by letting sort do the work for me, but similarly to my last response, it probably makes sense for me to add it to my schema as a sanity check if nothing else. Thanks, Chris Derrell.Lipman wrote: > > Chris Jones <[EMAIL PROTECTED]> writes: > >> Derrell.Lipman wrote: > > > So to guarantee that the *strings* are unique, you need a UNIQUE index on > the > string field. The ROWID is the INTEGER PRIMARY KEY whether you specify a > different name for it or not, but that will not guarantee that each of > your > *strings* is unique. Only a UNIQUE index on the string field will do > that. > > As long as you understand this, no need to reply. > > Cheers, > > Derrell > > - > To unsubscribe, send email to [EMAIL PROTECTED] > - > > > -- View this message in context: http://www.nabble.com/Index-creation-on-huge-table-will-never-finish.-tf3444218.html#a9626741 Sent from the SQLite mailing list archive at Nabble.com. - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
--- [EMAIL PROTECTED] wrote: > Gerry Snyder <[EMAIL PROTECTED]> wrote: > > Chris Jones wrote: > > > Hi all, > > > > > > I have a very simple schema. I need to assign a unique identifier to a > > > large collection of strings, each at most 80-bytes, although typically > > > shorter. > > > > > > The problem is I have 112 million of them. > > > > Maybe you could start by breaking the data into 8 equal groups and make > > a table of each group. Then merge the original groups pairwise, then > > merge those 4 groups, and finally the 2 semifinal groups (kinda like > > March Madness, come to think of it). Since each merging will be of > > already sorted/indexed data, it might save a lot of time. > > > > This is the right idea. > > The problem is that your working set is bigger than your cache > which is causing thrashing. I suggest a solution like this: > > Add entries to table ONE until the table and its unique index get > so big that they no longer fit in cache all at once. Then > transfer ONE into TWO like this: > >INSERT INTO two SELECT * FROM one ORDER BY unique_column; > > The ORDER BY is important here. > > Do the above a time or two until TWO is signficantly larger than ONE. > Then do the same into TWO_B. Later combine TWO and TWO_B into THREE: > >INSERT INTO three SELECT * FROM two ORDER BY unique_column; >INSERT INTO three SELECT * FROM two_b ORDER BY unique_column; > > Repeat as necessary for FOUR, FIVE, SIX and so forth. I've tried something like the algorithm you've proposed and it's much slower than pre-sorting all the data prior to bulk insert. It may have something to do with the cost of multiple repeated inserts for each original row. Here's a different attempt at speeding up bulk insert following your suggestion from Ticket 2075: http://www.mail-archive.com/sqlite-users%40sqlite.org/msg22143.html Any suggestions to speed it up are welcome. Sample insert speed test included. Don't get soaked. Take a quick peek at the forecast with the Yahoo! Search weather shortcut. http://tools.search.yahoo.com/shortcuts/#loc_weather - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones wrote: I probably should have made this more explicit, but in sqlite, every row has a unique identifier named rowid, which exists even if it isn't explicity declared in the schema, and I was depending on that. If you declare a PRIMARY KEY, then this replaces rowid. A tiny correction: a column will not replace rowid unless it is INTEGER PRIMARY KEY. From the web page: Specifying a PRIMARY KEY normally just creates a UNIQUE index on the corresponding columns. However, if primary key is on a single column that has datatype INTEGER, then that column is used internally as the actual key of the B-Tree for the table. Gerry - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones <[EMAIL PROTECTED]> writes: > Derrell.Lipman wrote: >> >> Chris Jones <[EMAIL PROTECTED]> writes: >> >> I don't think that your original solution solves that problem either. You >> first posted this schema: >> >>> My schema looks as follows: >>> >>> CREATE TABLE rawfen ( fen VARCHAR(80) ); >>> CREATE INDEX rawfen_idx_fen ON rawfen(fen); >> >> but you didn't indicate that rawfen_idx_fen is a UNIQUE INDEX so it won't >> complain if there are duplicate strings. To accomplish this (but not >> solving >> your timing issue), you need this: >> >> CREATE TABLE rawfen ( fen VARCHAR(80) ); >> CREATE UNIQUE INDEX rawfen_idx_fen ON rawfen(fen); >> > > I probably should have made this more explicit, but in sqlite, every row has > a unique identifier named rowid, which exists even if it isn't explicity > declared in the schema, and I was depending on that. If you declare a > PRIMARY KEY, then this replaces rowid. > > Of course, it's probably better practice to explicitly declare the primary > key, but anyway, that's where I was going with it. Those are two separate issues. Your declared problem was: > I don't think that solves my problem. Sure, it guarantees that the IDs are > unique, but not the strings. So to guarantee that the *strings* are unique, you need a UNIQUE index on the string field. The ROWID is the INTEGER PRIMARY KEY whether you specify a different name for it or not, but that will not guarantee that each of your *strings* is unique. Only a UNIQUE index on the string field will do that. As long as you understand this, no need to reply. Cheers, Derrell - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones wrote: So, I did a "sort -u -S 1800M fenout.txt > fenoutsort.txt" The sort took about 45 minutes, which is acceptable for me (it was much longer without the -S option to tell it to make use of more memory), and then loading the table was very efficient. Inserting all the rows into my table in sorted order took only 18 minutes. The sort command is more sophisticated than its usage info would lead one (at least me) to believe. After I saw your post I checked the sort source and found that it does all the splitting, sorting, and merging that I had suggested internally when given a file bigger than its allowed buffer space. No need for any scripts at all. I'm glad you found a suitable solution. Dennis Cote - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
--- Dennis Cote <[EMAIL PROTECTED]> wrote: > You could also improve the locality in the database file further by > running a vacuum command after it has been created. this will move the > pages around so that the page of the table are contiguous and so are the > pages of the index, rather than having them interspersed with each other. In SQLite 3.3.13 and earlier, VACUUM does indeed intermix the table pages with the index pages leading to fragmentation. It's also quite slow on large tables with multiple indexes unless you use a large cache. This patch in CVS address these problems: http://www.sqlite.org/cvstrac/chngview?cn=3643 See also: Changes to support fragmentation analysis in sqlite3_analyzer. http://www.sqlite.org/cvstrac/chngview?cn=3634 Need Mail bonding? Go to the Yahoo! Mail Q for great tips from Yahoo! Answers users. http://answers.yahoo.com/dir/?link=list=396546091 - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
A fast technique to achieve your objective is to perform what I believe is called a "monkey puzzle" sort. The data is not moved, instead an array of descriptors to each element is sorted. The output is realized by scanning the list of descriptors and picking up the associated record from the input list. Using a modest machine your application should run in less than ten minutes using that method. One way we use it is as a first stage in building a B-Tree index rapidly. Chris Jones wrote: Thanks everyone for your feedback. I ended up doing a presort on the data, and then adding the data in order. At first I was a little concerned about how I was going to implement an external sort on a data set that huge, and realized that the unix "sort" command can handle large files, and in fact does it pretty efficiently. So, I did a "sort -u -S 1800M fenout.txt > fenoutsort.txt" The sort took about 45 minutes, which is acceptable for me (it was much longer without the -S option to tell it to make use of more memory), and then loading the table was very efficient. Inserting all the rows into my table in sorted order took only 18 minutes. So, all in all, I can now load the table in just about an hour, which is great news for me. Thanks! Chris - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
An issue with cache is cache shadowing, the churning as data is copied from one cache to another to another. An example is the speed-up achieved on network accesses by using sendfile or TransmitFile and bypassing up to four levels of buffering for a message being despatched to a network interface using send or similar. Another is opening a file using open and writing to it with write and reading with read when you are transferring buffers full of data to or from a file. You avoid the extra level of buffer copying inherent in fopen. If you dispense with the reads and writes and access the virtual memory directly you get a further win. A modern OS uses main memory as a file system cache by virtue of its virtual memory capability. If you take advantage of that your applications will run faster. P Kishor wrote: On 3/22/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: "P Kishor" <[EMAIL PROTECTED]> wrote: > Richard, > > On 3/22/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > ... > > The problem is that your working set is bigger than your cache > > which is causing thrashing. I suggest a solution like this: > > > > Add entries to table ONE until the table and its unique index get > > so big that they no longer fit in cache all at once. > > What is this "cache" you talk about? Is this the hard disk cache, or > some kind of OS-level cache, or a SQLite level cache? All of the above. > How can I find > the size of this, and is this cache user-adjustable? If yes, how? > On Linux, the "top" command will show you how much OS disk cache you are currently using. The hard disk cache is usually small enough that it can be neglected. The OS cache is also usually much larger than SQLite's own internal cache. You might run an experiment where you start inserting and timing each insert, then switch to a different table when the insert speed drops below a threshold. ok. I was hoping for something magical like 'showme_cache_size'. This sounds more voodoo-ish, but do-able. In any case, it doesn't seem that this OS cache is user-adjustable, at least not without pipe wrenches and WD-40. On a note more related to the OP, even if one could adjust the cache to the working set (and that would be a moving target -- either make the cache gigantic, or keep on querying the relative sizes of the two and somehow keep on adjusting the cache, the time taken to determine uniqueness in a bigger working set will keep on increasing as the working set itself keeps on increasing... the curve won't be flat. - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
On 3/22/07, Martin Jenkins <[EMAIL PROTECTED]> wrote: Chris Jones wrote: > realized that the unix "sort" If I'd known you were on 'nix I'd have suggested using 'sort' and/or 'md5sum' about 12 hours ago. ;) Mac/Unix person meself, but the Windows XP sort is pretty darn good as well. -- Puneet Kishor http://punkish.eidesis.org/ Nelson Inst. for Env. Studies, UW-Madison http://www.nelson.wisc.edu/ Open Source Geospatial Foundation http://www.osgeo.org/education/ - collaborate, communicate, compete = - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones wrote: realized that the unix "sort" If I'd known you were on 'nix I'd have suggested using 'sort' and/or 'md5sum' about 12 hours ago. ;) Martin - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
You could sort the table then perform a merge which removes duplicates. Chris Jones wrote: I don't think that solves my problem. Sure, it guarantees that the IDs are unique, but not the strings. My whole goal is to be able to create a unique identifier for each string, in such a way that I dont have the same string listed twice, with different identifiers. In your solution, there is no way to lookup a string to see if it already exists, since there is no index on the string. Thanks, Chris - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Thanks everyone for your feedback. I ended up doing a presort on the data, and then adding the data in order. At first I was a little concerned about how I was going to implement an external sort on a data set that huge, and realized that the unix "sort" command can handle large files, and in fact does it pretty efficiently. So, I did a "sort -u -S 1800M fenout.txt > fenoutsort.txt" The sort took about 45 minutes, which is acceptable for me (it was much longer without the -S option to tell it to make use of more memory), and then loading the table was very efficient. Inserting all the rows into my table in sorted order took only 18 minutes. So, all in all, I can now load the table in just about an hour, which is great news for me. Thanks! Chris -- View this message in context: http://www.nabble.com/Index-creation-on-huge-table-will-never-finish.-tf3444218.html#a9618709 Sent from the SQLite mailing list archive at Nabble.com. - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones wrote: I've read elsewhere that this is a data locality issue, which certainly makes sense. And in those threads, a suggestion has been made to insert in sorted order. But it's unclear to me exactly what the sorting function would need to be - it's likely my sorting function (say strcmp()) wouldn't match what sqlite expects. It's also a little unclear if I can even sort this many keys externally with any ease. After all, that's why I'm trying to use a database. Chris, You are correct that this is a locality of reference issue. Using a presorted list speeds insertions because all the index insertions to be made at end of the index. To make this run faster you need prepare a presorted list. The question is how best to do that with a large file. You didn't say what OS you are running, but these days most basic unix commands are available on Windows as well. I ams assuming you have your list of 112M lines in a text file. You can prepare a large sorted file pretty efficiently using a merge sort, since it reads the input files in one pass. The trick is the input files must be presorted. So you need to split your original data file into many smaller files, sort each in memory (where locality isn't a problem as long as the data fits in real memory), and then merge the individual sorted files. This is best done is a script. You can use the split command to split you text file into many separate files. If you split your file into chunks of 1M lines you will have 112 files to sort and merge. Each file will be less than 80 MB, so it should be in memory sortable. split -l 100 infile part This will generate 112 files names partaa, partab, partac ... Now you need to sort each of these files. Note the sort command can also eliminate any duplicate lines in your files (the -u for unique option). parts=`ls part*` for f in $parts do sort -o $f -u $f done Now you have 112 sorted files that need to be merged. mv partaa sorted parts=`ls part*` for f in $parts do sort -m -o sorted sorted $f done You now have a single large sorted file that contains unique strings. You should be able to insert the strings from this file into your sqlite database much faster than you could before. If you want you can get fancier with the script and merge pairs of sorted files into increasingly larger files until you have a single file. I'll leave that as an exercise for the reader. You could also improve the locality in the database file further by running a vacuum command after it has been created. this will move the pages around so that the page of the table are contiguous and so are the pages of the index, rather than having them interspersed with each other. HTH Dennis Cote A shell script that takes your raw text file as an argument. #!/bin/sh split -l 100 $1 part parts=`ls part*` for f in $parts do sort -o $f -u $f done mv partaa sorted parts=`ls part*` for f in $parts do sort -m -o sorted sorted $f done - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
"P Kishor" <[EMAIL PROTECTED]> wrote: > Richard, > > On 3/22/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: > ... > > The problem is that your working set is bigger than your cache > > which is causing thrashing. I suggest a solution like this: > > > > Add entries to table ONE until the table and its unique index get > > so big that they no longer fit in cache all at once. > > What is this "cache" you talk about? Is this the hard disk cache, or > some kind of OS-level cache, or a SQLite level cache? All of the above. > How can I find > the size of this, and is this cache user-adjustable? If yes, how? > On Linux, the "top" command will show you how much OS disk cache you are currently using. The hard disk cache is usually small enough that it can be neglected. The OS cache is also usually much larger than SQLite's own internal cache. You might run an experiment where you start inserting and timing each insert, then switch to a different table when the insert speed drops below a threshold. -- D. Richard Hipp <[EMAIL PROTECTED]> - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
At 04:47 22/03/2007, you wrote: I don't think that solves my problem. Sure, it guarantees that the IDs are unique, but not the strings. My whole goal is to be able to create a unique identifier for each string, in such a way that I dont have the same string listed twice, with different identifiers. In your solution, there is no way to lookup a string to see if it already exists, since there is no index on the string. Thanks, Chris So you have a file with data, a large collection of strings 112 millions, each at most 80-bytes, although typically shorter. How do you manage repeated data? Replace? First In? Modify string to be unique? You want put them in a sqlite3 database, but each string must be only once. The problem i see here is if you have a data file with repeated strings or not. I think that a grep or a perl script can help you a lot cleaning your data first. Then import to database will be fast. HTH -- "Hemos encontrado al enemigo y somos nosotros" - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Richard, On 3/22/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: .. The problem is that your working set is bigger than your cache which is causing thrashing. I suggest a solution like this: Add entries to table ONE until the table and its unique index get so big that they no longer fit in cache all at once. What is this "cache" you talk about? Is this the hard disk cache, or some kind of OS-level cache, or a SQLite level cache? How can I find the size of this, and is this cache user-adjustable? If yes, how? -- Puneet Kishor http://punkish.eidesis.org/ Nelson Inst. for Env. Studies, UW-Madison http://www.nelson.wisc.edu/ Open Source Geospatial Foundation http://www.osgeo.org/education/ - collaborate, communicate, compete = - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Gerry Snyder <[EMAIL PROTECTED]> wrote: > Chris Jones wrote: > Hi all, > I have a very simple schema. I need to assign a unique identifier > to a large collection of strings, each at most 80-bytes, although > typically shorter. Would it help to hash the strings, then save them in the DB, checking the hash instead of the string for duplication? You might still get duplicates, depending on what hashing algorithm you use, but those should be relatively easy to find afterwards. Hashing could take a while, but that should be a linear time operation all by itself. - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Ion Silvestru <[EMAIL PROTECTED]> wrote: > > drh wrote: > > INSERT INTO two SELECT * FROM one ORDER BY unique_column; > > >The ORDER BY is important here. > > This is an excerpt from SQLite documentation: > > The second form of the INSERT statement takes it data from a SELECT statement. > The number of columns in the result of the SELECT must exactly match the > number > of columns in the table if no column list is specified, > or it must match the number of columns name in the column list. > A new entry is made in the table for every row of the SELECT result. > The SELECT may be simple or compound. > If the SELECT statement has an ORDER BY clause, the ORDER BY is ignored. > ^ > > Question: ORDER BY is ignored or not ? > You are looking at obsolete documentation. See http://www.sqlite.org/cvstrac/tktview?tn=1923 http://www.sqlite.org/cvstrac/chngview?cn=3356 -- D. Richard Hipp <[EMAIL PROTECTED]> - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones <[EMAIL PROTECTED]> writes: > I don't think that solves my problem. Sure, it guarantees that the IDs are > unique, but not the strings. > > My whole goal is to be able to create a unique identifier for each string, > in such a way that I dont have the same string listed twice, with different > identifiers. I don't think that your original solution solves that problem either. You first posted this schema: > My schema looks as follows: > > CREATE TABLE rawfen ( fen VARCHAR(80) ); > CREATE INDEX rawfen_idx_fen ON rawfen(fen); but you didn't indicate that rawfen_idx_fen is a UNIQUE INDEX so it won't complain if there are duplicate strings. To accomplish this (but not solving your timing issue), you need this: CREATE TABLE rawfen ( fen VARCHAR(80) ); CREATE UNIQUE INDEX rawfen_idx_fen ON rawfen(fen); or, more concisely, CREATE TABLE rawfen ( fen VARCHAR(80) UNIQUE); As previously stated, you can guarantee a unique id for each string with CREATE TABLE rawfen (id INTEGER PRIMARY KEY, fen VARCHAR(80) UNIQUE); Cheers, Derrell - To unsubscribe, send email to [EMAIL PROTECTED] -
Re[2]: [sqlite] Index creation on huge table will never finish.
> drh wrote: > INSERT INTO two SELECT * FROM one ORDER BY unique_column; >The ORDER BY is important here. This is an excerpt from SQLite documentation: The second form of the INSERT statement takes it data from a SELECT statement. The number of columns in the result of the SELECT must exactly match the number of columns in the table if no column list is specified, or it must match the number of columns name in the column list. A new entry is made in the table for every row of the SELECT result. The SELECT may be simple or compound. If the SELECT statement has an ORDER BY clause, the ORDER BY is ignored. ^ Question: ORDER BY is ignored or not ? - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Gerry Snyder <[EMAIL PROTECTED]> wrote: > Chris Jones wrote: > > Hi all, > > > > I have a very simple schema. I need to assign a unique identifier to a > > large collection of strings, each at most 80-bytes, although typically > > shorter. > > > > The problem is I have 112 million of them. > > Maybe you could start by breaking the data into 8 equal groups and make > a table of each group. Then merge the original groups pairwise, then > merge those 4 groups, and finally the 2 semifinal groups (kinda like > March Madness, come to think of it). Since each merging will be of > already sorted/indexed data, it might save a lot of time. > This is the right idea. The problem is that your working set is bigger than your cache which is causing thrashing. I suggest a solution like this: Add entries to table ONE until the table and its unique index get so big that they no longer fit in cache all at once. Then transfer ONE into TWO like this: INSERT INTO two SELECT * FROM one ORDER BY unique_column; The ORDER BY is important here. Do the above a time or two until TWO is signficantly larger than ONE. Then do the same into TWO_B. Later combine TWO and TWO_B into THREE: INSERT INTO three SELECT * FROM two ORDER BY unique_column; INSERT INTO three SELECT * FROM two_b ORDER BY unique_column; Repeat as necessary for FOUR, FIVE, SIX and so forth. -- D. Richard Hipp <[EMAIL PROTECTED]> - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
Chris Jones wrote: Hi all, I have a very simple schema. I need to assign a unique identifier to a large collection of strings, each at most 80-bytes, although typically shorter. The problem is I have 112 million of them. Maybe you could start by breaking the data into 8 equal groups and make a table of each group. Then merge the original groups pairwise, then merge those 4 groups, and finally the 2 semifinal groups (kinda like March Madness, come to think of it). Since each merging will be of already sorted/indexed data, it might save a lot of time. Or perhaps do a block sort based on the first character of the string (or nth char if most of the first chars are the same) and have a bunch of smaller tables with that character as (part of) the table name. The global unique identifier could be the character concatenated with the rowid in its table. HTH, Gerry - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
You stated in your OP I need to assign a unique identifier to a large collection of strings Obviously I misunderstood that to mean you wanted the strings tagged uniquely, not that the strings were unique. In your case, it seems then, you will have to put up with checking each string, and as the db gets bigger, that task will take longer time. Maybe someone else has other ideas... On 3/21/07, Chris Jones <[EMAIL PROTECTED]> wrote: I don't think that solves my problem. Sure, it guarantees that the IDs are unique, but not the strings. My whole goal is to be able to create a unique identifier for each string, in such a way that I dont have the same string listed twice, with different identifiers. In your solution, there is no way to lookup a string to see if it already exists, since there is no index on the string. Thanks, Chris -- View this message in context: http://www.nabble.com/Index-creation-on-huge-table-will-never-finish.-tf3444218.html#a9607996 Sent from the SQLite mailing list archive at Nabble.com. - To unsubscribe, send email to [EMAIL PROTECTED] - -- Puneet Kishor http://punkish.eidesis.org/ Nelson Inst. for Env. Studies, UW-Madison http://www.nelson.wisc.edu/ Open Source Geospatial Foundation http://www.osgeo.org/education/ - collaborate, communicate, compete = - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
I don't think that solves my problem. Sure, it guarantees that the IDs are unique, but not the strings. My whole goal is to be able to create a unique identifier for each string, in such a way that I dont have the same string listed twice, with different identifiers. In your solution, there is no way to lookup a string to see if it already exists, since there is no index on the string. Thanks, Chris -- View this message in context: http://www.nabble.com/Index-creation-on-huge-table-will-never-finish.-tf3444218.html#a9607996 Sent from the SQLite mailing list archive at Nabble.com. - To unsubscribe, send email to [EMAIL PROTECTED] -
Re: [sqlite] Index creation on huge table will never finish.
On 3/21/07, Chris Jones <[EMAIL PROTECTED]> wrote: Hi all, I have a very simple schema. I need to assign a unique identifier to a large collection of strings, each at most 80-bytes, although typically shorter. The problem is I have 112 million of them. My schema looks as follows: CREATE TABLE rawfen ( fen VARCHAR(80) ); CREATE INDEX rawfen_idx_fen ON rawfen(fen); Unforuntately, data loading this table takes virtually forever. After 24 hours, its not finished, and that is inserting rows in transactions of 100,000 rows per transaction. I tried dropping the index, and building it after row insertion. That has two problems. First, since I have no index, I can't guarantee string uniqueness (and I'd like to). Second, it still doesn't solve my speed problem. Insertion without the index takes a little over an hour, but the index creation never finishes. Well, I gave it 6 hours and it was unclear if it was making any progress. .. It is the uniqueness that is killing it. Doesn't matter how you do it. As more your bucket becomes bigger, it has to do more work to determine if the string is unique or not. I tried a little test. First with CREATE TABLE tbl (fen PRIMARY KEY) I got results that looked like so, committing every 100,000,... 10: 9 wallclock secs ( 7.36 usr + 1.63 sys = 8.99 CPU) 20: 12 wallclock secs ( 8.06 usr + 2.87 sys = 10.93 CPU) 30: 18 wallclock secs ( 8.27 usr + 3.29 sys = 11.56 CPU) 40: 14 wallclock secs ( 8.32 usr + 3.50 sys = 11.82 CPU) 50: 14 wallclock secs ( 8.45 usr + 3.71 sys = 12.16 CPU) 60: 15 wallclock secs ( 8.49 usr + 3.89 sys = 12.38 CPU) 70: 16 wallclock secs ( 8.68 usr + 4.15 sys = 12.83 CPU) 80: 15 wallclock secs ( 8.61 usr + 4.16 sys = 12.77 CPU) 90: 16 wallclock secs ( 8.69 usr + 4.29 sys = 12.98 CPU) 100: 17 wallclock secs ( 8.65 usr + 4.38 sys = 13.03 CPU) Then I tried with CREATE TABLE tbl (fen) It was a more flat curve like so... 10: 5 wallclock secs ( 5.19 usr + 0.09 sys = 5.28 CPU) 20: 6 wallclock secs ( 5.23 usr + 0.09 sys = 5.32 CPU) 30: 5 wallclock secs ( 5.24 usr + 0.09 sys = 5.33 CPU) 40: 6 wallclock secs ( 5.23 usr + 0.10 sys = 5.33 CPU) 50: 5 wallclock secs ( 5.22 usr + 0.09 sys = 5.31 CPU) 60: 5 wallclock secs ( 5.24 usr + 0.10 sys = 5.34 CPU) 70: 5 wallclock secs ( 5.23 usr + 0.09 sys = 5.32 CPU) 80: 6 wallclock secs ( 5.24 usr + 0.10 sys = 5.34 CPU) 90: 6 wallclock secs ( 5.26 usr + 0.10 sys = 5.36 CPU) 100: 5 wallclock secs ( 5.24 usr + 0.10 sys = 5.34 CPU) Well, since you want to tag each string with a unique id, how about CREATE TABLE tbl (id INTEGER PRIMARY KEY, fen) I get a flat curve again... 10: 6 wallclock secs ( 5.22 usr + 0.10 sys = 5.32 CPU) 20: 5 wallclock secs ( 5.24 usr + 0.09 sys = 5.33 CPU) 30: 6 wallclock secs ( 5.26 usr + 0.10 sys = 5.36 CPU) 40: 5 wallclock secs ( 5.26 usr + 0.10 sys = 5.36 CPU) 50: 5 wallclock secs ( 5.27 usr + 0.11 sys = 5.38 CPU) 60: 6 wallclock secs ( 5.27 usr + 0.10 sys = 5.37 CPU) 70: 6 wallclock secs ( 5.27 usr + 0.09 sys = 5.36 CPU) 80: 6 wallclock secs ( 5.27 usr + 0.10 sys = 5.37 CPU) 90: 6 wallclock secs ( 5.26 usr + 0.10 sys = 5.36 CPU) 100: 5 wallclock secs ( 5.26 usr + 0.10 sys = 5.36 CPU) Problem solved... id is your unique tag. It will be done in about 1.5 hours on my MacBook Pro. -- Puneet Kishor http://punkish.eidesis.org/ Nelson Inst. for Env. Studies, UW-Madison http://www.nelson.wisc.edu/ Open Source Geospatial Foundation http://www.osgeo.org/education/ - collaborate, communicate, compete = - To unsubscribe, send email to [EMAIL PROTECTED] -
[sqlite] Index creation on huge table will never finish.
Hi all, I have a very simple schema. I need to assign a unique identifier to a large collection of strings, each at most 80-bytes, although typically shorter. The problem is I have 112 million of them. My schema looks as follows: CREATE TABLE rawfen ( fen VARCHAR(80) ); CREATE INDEX rawfen_idx_fen ON rawfen(fen); Unforuntately, data loading this table takes virtually forever. After 24 hours, its not finished, and that is inserting rows in transactions of 100,000 rows per transaction. I tried dropping the index, and building it after row insertion. That has two problems. First, since I have no index, I can't guarantee string uniqueness (and I'd like to). Second, it still doesn't solve my speed problem. Insertion without the index takes a little over an hour, but the index creation never finishes. Well, I gave it 6 hours and it was unclear if it was making any progress. I've read elsewhere that this is a data locality issue, which certainly makes sense. And in those threads, a suggestion has been made to insert in sorted order. But it's unclear to me exactly what the sorting function would need to be - it's likely my sorting function (say strcmp()) wouldn't match what sqlite expects. It's also a little unclear if I can even sort this many keys externally with any ease. After all, that's why I'm trying to use a database. Last, is this something that is likely to affect me if I port over to another database? Do others, say mySQL for instance, handle this better? Does anyone have any suggestions?Upgrading my server isn't an option - it already has 4Gig of memory and very fast disks in a RAID-5 configuration (with a 512MB cache on the raid controller). Thanks! -- View this message in context: http://www.nabble.com/Index-creation-on-huge-table-will-never-finish.-tf3444218.html#a9604705 Sent from the SQLite mailing list archive at Nabble.com. - To unsubscribe, send email to [EMAIL PROTECTED] -
