Intermedia Performance Benchmarks anyone ?
Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Martin Kendall INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).
RE: Intermedia Performance Benchmarks anyone ?
Title: RE: Intermedia Performance Benchmarks anyone ? Hi Martin, I've had to implement intermedia in the past. with 1 million records intermedia will do the job in an acceptable amount of time. However, when testing it I found that the more advanced features I used (fuzzy, etc.) the worse performance became. It works well if you are doing a straightforward search. I also found there were problems with creating the library on HP/UX (part of the install). I had to create it manually. There are also tricks for indexing more than one column in one index. The documentation says you can't, but it can be done. I know there are people on the list that have done this. I didn't run any formal benchmarks. We had indexed ~6million records on long fields (like description) and performance was directly related to how selective the keywords were. Indexing time - If I remember right, creating one index on one of these description columns ran in ~2 hours. I also batched my updates. (I didn't run ctxsrv) The index ended up being ~1.5GB. Hope this helps and let me know if you have any other questions, I'll try and answer them. Lisa Koivu Oracle Database Administrator Fairfield Resorts, Inc. 954-935-4117 -Original Message- From: Martin Kendall [SMTP:[EMAIL PROTECTED]] Sent: Thursday, October 04, 2001 11:00 AM To: Multiple recipients of list ORACLE-L Subject: Intermedia Performance Benchmarks anyone ? Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Martin Kendall INET: [EMAIL PROTECTED] Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051 San Diego, California -- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).
RE: Intermedia Performance Benchmarks anyone ?
Martin, We use interMedia Text to index and query up to about 10-15 million CLOB documents (up to 5KB each). We're on 8.1.6.0.0 under Win2k - 2 550MHz CPUs, 2GB RAM, 18 36GB drives. Because a domain index cannot be partitioned, we have the documents spread across 5 tables (on 6 drives). One is a 2 partition table (each partition on its own drive) containing the current two months of docs, the other 4 hold the 4 prior months' docs. We can query the entire 6 months of docs via a Union View on them - even Contains() queries work fine on this view. When we add a new month's partition, the prior month's partition gets turned into a table (segment exchange). The interMedia Text indexes on the partitioned table and the new prior month are rebuilt. Lately we've been getting about 3.5 million docs/month and the index rebuild takes about 7 hours - that's 7 hrs. for the index on the prior month and 7 more hours for the index on the partitioned table, which only contains one month of docs at that point. Since we're adding docs every day, we sync the interMedia index every morning. Last night we added about 200,000 docs and it took about 3 hours for the index to resync. We don't use ctxsrv, but use CTX_DDL.Sync_Index. When we get over about 4.5 million docs in a table, the resync really slows down. The in-memory part still happens at about 150 docs/sec, but when interMedia writes to disk it slows down a bunch. What took 3 hours today will take 10 hours in a couple of weeks. That's why I plan on spreading the DR$<>$I segment across multiple drives by spreading the datafiles of its tablespace across those drives. BTW, that brings up some performance points - be sure you cache the DR$<>$R segment (use CACHE not CACHE READS, due to bugs in Oracle): Alter Table DR$$R Modify LOB (Data) (Cache) ; Also ensure that your LOBs are out-of-line and stored in their own segment(s) on drive(s) separate from the "regular" data. Make sure that your I_TABLE_CLAUSE, R_TABLE_CLAUSE, and I_INDEX_CLAUSE all specify tablespaces on their own drives to spread the I/O out even further. We're getting 2GB more RAM on a new server, so I plan on caching the 900MB DR$<>$X segment, which is the index on the DR$<>$I token table. I've learned a lot about how interMedia Text processes different kinds of queries by watching disk I/O on Win2k's Performance Monitor while I issue various "flavors". Our folks use lots of complex query terms with heavy use of the Stemmer. I've gotten them to switch from using tons of ORs to using the Equivalence operator and we're getting much better results using NEAR than simple ANDs. Performance is very good, with CONTAINS queries returning results in less than a second for terms that are rare in the docs, up to a minute for terms that are common in lots (e.g. hundreds of thousands) of docs. If you're going to do synonym searches, you'd better start looking for a good thesaurus - the one Oracle ships is pretty limited. We've not found a good one for the technical lingo our docs contain, so we don't do ABOUT queries at this time. Get familiar with CTX_Query.Explain, it will help you understand things like what the Stemmer *really* does and how complex queries are parsed. Hope this helps. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Kendall Sent: Thursday, October 04, 2001 10:00 AM To: Multiple recipients of list ORACLE-L Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack C. Applewhite INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).
RE: Intermedia Performance Benchmarks anyone ?
*excellent* post. thanks. Anyone out there put the indexes and tables on solid state disk? They have ssd up to about 10G and higher, I hearjust curious, not trying to invoke a global listserv discussion on how it "can't work" or "wouldn't be worth it, especially on microsoft platforms", etc. It would be neat to hear about an InterMedia indexing miracle. This really neat tool just sounds WAAY to slow to scale at this point, which answers a pet question of mine. (Something like "Why do services like 'Ask Jeeves' suck so hard?") In Love and Peas, etc. -Original Message- Sent: Thursday, October 04, 2001 5:47 PM To: Multiple recipients of list ORACLE-L Martin, We use interMedia Text to index and query up to about 10-15 million CLOB documents (up to 5KB each). We're on 8.1.6.0.0 under Win2k - 2 550MHz CPUs, 2GB RAM, 18 36GB drives. Because a domain index cannot be partitioned, we have the documents spread across 5 tables (on 6 drives). One is a 2 partition table (each partition on its own drive) containing the current two months of docs, the other 4 hold the 4 prior months' docs. We can query the entire 6 months of docs via a Union View on them - even Contains() queries work fine on this view. When we add a new month's partition, the prior month's partition gets turned into a table (segment exchange). The interMedia Text indexes on the partitioned table and the new prior month are rebuilt. Lately we've been getting about 3.5 million docs/month and the index rebuild takes about 7 hours - that's 7 hrs. for the index on the prior month and 7 more hours for the index on the partitioned table, which only contains one month of docs at that point. Since we're adding docs every day, we sync the interMedia index every morning. Last night we added about 200,000 docs and it took about 3 hours for the index to resync. We don't use ctxsrv, but use CTX_DDL.Sync_Index. When we get over about 4.5 million docs in a table, the resync really slows down. The in-memory part still happens at about 150 docs/sec, but when interMedia writes to disk it slows down a bunch. What took 3 hours today will take 10 hours in a couple of weeks. That's why I plan on spreading the DR$<>$I segment across multiple drives by spreading the datafiles of its tablespace across those drives. BTW, that brings up some performance points - be sure you cache the DR$<>$R segment (use CACHE not CACHE READS, due to bugs in Oracle): Alter Table DR$$R Modify LOB (Data) (Cache) ; Also ensure that your LOBs are out-of-line and stored in their own segment(s) on drive(s) separate from the "regular" data. Make sure that your I_TABLE_CLAUSE, R_TABLE_CLAUSE, and I_INDEX_CLAUSE all specify tablespaces on their own drives to spread the I/O out even further. We're getting 2GB more RAM on a new server, so I plan on caching the 900MB DR$<>$X segment, which is the index on the DR$<>$I token table. I've learned a lot about how interMedia Text processes different kinds of queries by watching disk I/O on Win2k's Performance Monitor while I issue various "flavors". Our folks use lots of complex query terms with heavy use of the Stemmer. I've gotten them to switch from using tons of ORs to using the Equivalence operator and we're getting much better results using NEAR than simple ANDs. Performance is very good, with CONTAINS queries returning results in less than a second for terms that are rare in the docs, up to a minute for terms that are common in lots (e.g. hundreds of thousands) of docs. If you're going to do synonym searches, you'd better start looking for a good thesaurus - the one Oracle ships is pretty limited. We've not found a good one for the technical lingo our docs contain, so we don't do ABOUT queries at this time. Get familiar with CTX_Query.Explain, it will help you understand things like what the Stemmer *really* does and how complex queries are parsed. Hope this helps. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Kendall Sent: Thursday, October 04, 2001 10:00 AM To: Multiple recipients of list ORACLE-L Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack
RE: Intermedia Performance Benchmarks anyone ?
Ross, I disagree that interMedia Text is "way too slow to scale". Our experience has convinced us that I/O bottlenecks are the main performance killers with large interMedia Text indexes. The problem is that it takes some experience to find out how this "special" kind of index is structured (6 or 8 separate table and index segments per "index") and how it behaves. As usual, the Oracle docs are pitifully inadequate - you've gotta search through TechNet and MetaLink for details and bug workarounds (like CACHE instead of CACHE READS for DR$<>$R). Caching the DR$<>$R segment helped immensely and I can see that when pieces of the DR$<>$X index are cached, queries with terms in those pieces are lightening fast. I am betting that when I spread the DR$<>$I table across multiple drives, instead of the single drive ours is currently on, we'll see much better performance of NEAR queries (which depend on the word position info. there), as well as faster index resyncs. In 9i Domain Indexes become partitionable, so I'm looking forward (in about a year - experiences with 6.0, 7.0, 8.0 and 8.1.5 have made me wary) to putting our 6 (or more) months of docs into one partitoned table. There may be other I/O distributing kinds of enhancements by then, as well. For sure I'll have explored every trick I can think of! ;-) With more drives and a bit more RAM, I think we can handle 10 million docs per month (60 million total online), even on our lil' ol' Win2k box. That's just x3 to x4 of what we do now. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Sent: Thursday, October 04, 2001 5:36 PM To: Multiple recipients of list ORACLE-L *excellent* post. thanks. Anyone out there put the indexes and tables on solid state disk? They have ssd up to about 10G and higher, I hearjust curious, not trying to invoke a global listserv discussion on how it "can't work" or "wouldn't be worth it, especially on microsoft platforms", etc. It would be neat to hear about an InterMedia indexing miracle. This really neat tool just sounds WAAY to slow to scale at this point, which answers a pet question of mine. (Something like "Why do services like 'Ask Jeeves' suck so hard?") In Love and Peas, etc. -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack C. Applewhite INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).
RE: Intermedia Performance Benchmarks anyone ?
I looked into SSD, they are like $15-17k / gb. Very expensive. And with Oracle buffering, I would expect the performance wouldn't be huge. There were some solutions that were like $2500 for a 1 gb, but they were not sharable between machines in a cluster. -Original Message- To: Multiple recipients of list ORACLE-L Sent: 10/4/01 6:35 PM *excellent* post. thanks. Anyone out there put the indexes and tables on solid state disk? They have ssd up to about 10G and higher, I hearjust curious, not trying to invoke a global listserv discussion on how it "can't work" or "wouldn't be worth it, especially on microsoft platforms", etc. It would be neat to hear about an InterMedia indexing miracle. This really neat tool just sounds WAAY to slow to scale at this point, which answers a pet question of mine. (Something like "Why do services like 'Ask Jeeves' suck so hard?") In Love and Peas, etc. -Original Message- Sent: Thursday, October 04, 2001 5:47 PM To: Multiple recipients of list ORACLE-L Martin, We use interMedia Text to index and query up to about 10-15 million CLOB documents (up to 5KB each). We're on 8.1.6.0.0 under Win2k - 2 550MHz CPUs, 2GB RAM, 18 36GB drives. Because a domain index cannot be partitioned, we have the documents spread across 5 tables (on 6 drives). One is a 2 partition table (each partition on its own drive) containing the current two months of docs, the other 4 hold the 4 prior months' docs. We can query the entire 6 months of docs via a Union View on them - even Contains() queries work fine on this view. When we add a new month's partition, the prior month's partition gets turned into a table (segment exchange). The interMedia Text indexes on the partitioned table and the new prior month are rebuilt. Lately we've been getting about 3.5 million docs/month and the index rebuild takes about 7 hours - that's 7 hrs. for the index on the prior month and 7 more hours for the index on the partitioned table, which only contains one month of docs at that point. Since we're adding docs every day, we sync the interMedia index every morning. Last night we added about 200,000 docs and it took about 3 hours for the index to resync. We don't use ctxsrv, but use CTX_DDL.Sync_Index. When we get over about 4.5 million docs in a table, the resync really slows down. The in-memory part still happens at about 150 docs/sec, but when interMedia writes to disk it slows down a bunch. What took 3 hours today will take 10 hours in a couple of weeks. That's why I plan on spreading the DR$<>$I segment across multiple drives by spreading the datafiles of its tablespace across those drives. BTW, that brings up some performance points - be sure you cache the DR$<>$R segment (use CACHE not CACHE READS, due to bugs in Oracle): Alter Table DR$$R Modify LOB (Data) (Cache) ; Also ensure that your LOBs are out-of-line and stored in their own segment(s) on drive(s) separate from the "regular" data. Make sure that your I_TABLE_CLAUSE, R_TABLE_CLAUSE, and I_INDEX_CLAUSE all specify tablespaces on their own drives to spread the I/O out even further. We're getting 2GB more RAM on a new server, so I plan on caching the 900MB DR$<>$X segment, which is the index on the DR$<>$I token table. I've learned a lot about how interMedia Text processes different kinds of queries by watching disk I/O on Win2k's Performance Monitor while I issue various "flavors". Our folks use lots of complex query terms with heavy use of the Stemmer. I've gotten them to switch from using tons of ORs to using the Equivalence operator and we're getting much better results using NEAR than simple ANDs. Performance is very good, with CONTAINS queries returning results in less than a second for terms that are rare in the docs, up to a minute for terms that are common in lots (e.g. hundreds of thousands) of docs. If you're going to do synonym searches, you'd better start looking for a good thesaurus - the one Oracle ships is pretty limited. We've not found a good one for the technical lingo our docs contain, so we don't do ABOUT queries at this time. Get familiar with CTX_Query.Explain, it will help you understand things like what the Stemmer *really* does and how complex queries are parsed. Hope this helps. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Kendall Sent: Thursday, October 04, 2001 10:00 AM To: Multiple recipients of list ORACLE-L Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographica
RE: Intermedia Performance Benchmarks anyone ?
Jack, Thanks for your time on this. Most revealing and useful for what I have ahead of me Kind regards from the UK. Martin -Original Message- Sent: 04 October 2001 22:47 To: Multiple recipients of list ORACLE-L Martin, We use interMedia Text to index and query up to about 10-15 million CLOB documents (up to 5KB each). We're on 8.1.6.0.0 under Win2k - 2 550MHz CPUs, 2GB RAM, 18 36GB drives. Because a domain index cannot be partitioned, we have the documents spread across 5 tables (on 6 drives). One is a 2 partition table (each partition on its own drive) containing the current two months of docs, the other 4 hold the 4 prior months' docs. We can query the entire 6 months of docs via a Union View on them - even Contains() queries work fine on this view. When we add a new month's partition, the prior month's partition gets turned into a table (segment exchange). The interMedia Text indexes on the partitioned table and the new prior month are rebuilt. Lately we've been getting about 3.5 million docs/month and the index rebuild takes about 7 hours - that's 7 hrs. for the index on the prior month and 7 more hours for the index on the partitioned table, which only contains one month of docs at that point. Since we're adding docs every day, we sync the interMedia index every morning. Last night we added about 200,000 docs and it took about 3 hours for the index to resync. We don't use ctxsrv, but use CTX_DDL.Sync_Index. When we get over about 4.5 million docs in a table, the resync really slows down. The in-memory part still happens at about 150 docs/sec, but when interMedia writes to disk it slows down a bunch. What took 3 hours today will take 10 hours in a couple of weeks. That's why I plan on spreading the DR$<>$I segment across multiple drives by spreading the datafiles of its tablespace across those drives. BTW, that brings up some performance points - be sure you cache the DR$<>$R segment (use CACHE not CACHE READS, due to bugs in Oracle): Alter Table DR$$R Modify LOB (Data) (Cache) ; Also ensure that your LOBs are out-of-line and stored in their own segment(s) on drive(s) separate from the "regular" data. Make sure that your I_TABLE_CLAUSE, R_TABLE_CLAUSE, and I_INDEX_CLAUSE all specify tablespaces on their own drives to spread the I/O out even further. We're getting 2GB more RAM on a new server, so I plan on caching the 900MB DR$<>$X segment, which is the index on the DR$<>$I token table. I've learned a lot about how interMedia Text processes different kinds of queries by watching disk I/O on Win2k's Performance Monitor while I issue various "flavors". Our folks use lots of complex query terms with heavy use of the Stemmer. I've gotten them to switch from using tons of ORs to using the Equivalence operator and we're getting much better results using NEAR than simple ANDs. Performance is very good, with CONTAINS queries returning results in less than a second for terms that are rare in the docs, up to a minute for terms that are common in lots (e.g. hundreds of thousands) of docs. If you're going to do synonym searches, you'd better start looking for a good thesaurus - the one Oracle ships is pretty limited. We've not found a good one for the technical lingo our docs contain, so we don't do ABOUT queries at this time. Get familiar with CTX_Query.Explain, it will help you understand things like what the Stemmer *really* does and how complex queries are parsed. Hope this helps. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Kendall Sent: Thursday, October 04, 2001 10:00 AM To: Multiple recipients of list ORACLE-L Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack C. Applewhite INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name
RE: Intermedia Performance Benchmarks anyone ?
Excellent doc. I just wonder if there are only cache related bugs on using interMedia. Mario Alberto Ramos >>> [EMAIL PROTECTED] 04/10/01 16:47 >>> Martin, We use interMedia Text to index and query up to about 10-15 million CLOB documents (up to 5KB each). We're on 8.1.6.0.0 under Win2k - 2 550MHz CPUs, 2GB RAM, 18 36GB drives. Because a domain index cannot be partitioned, we have the documents spread across 5 tables (on 6 drives). One is a 2 partition table (each partition on its own drive) containing the current two months of docs, the other 4 hold the 4 prior months' docs. We can query the entire 6 months of docs via a Union View on them - even Contains() queries work fine on this view. When we add a new month's partition, the prior month's partition gets turned into a table (segment exchange). The interMedia Text indexes on the partitioned table and the new prior month are rebuilt. Lately we've been getting about 3.5 million docs/month and the index rebuild takes about 7 hours - that's 7 hrs. for the index on the prior month and 7 more hours for the index on the partitioned table, which only contains one month of docs at that point. Since we're adding docs every day, we sync the interMedia index every morning. Last night we added about 200,000 docs and it took about 3 hours for the index to resync. We don't use ctxsrv, but use CTX_DDL.Sync_Index. When we get over about 4.5 million docs in a table, the resync really slows down. The in-memory part still happens at about 150 docs/sec, but when interMedia writes to disk it slows down a bunch. What took 3 hours today will take 10 hours in a couple of weeks. That's why I plan on spreading the DR$<>$I segment across multiple drives by spreading the datafiles of its tablespace across those drives. BTW, that brings up some performance points - be sure you cache the DR$<>$R segment (use CACHE not CACHE READS, due to bugs in Oracle): Alter Table DR$$R Modify LOB (Data) (Cache) ; Also ensure that your LOBs are out-of-line and stored in their own segment(s) on drive(s) separate from the "regular" data. Make sure that your I_TABLE_CLAUSE, R_TABLE_CLAUSE, and I_INDEX_CLAUSE all specify tablespaces on their own drives to spread the I/O out even further. We're getting 2GB more RAM on a new server, so I plan on caching the 900MB DR$<>$X segment, which is the index on the DR$<>$I token table. I've learned a lot about how interMedia Text processes different kinds of queries by watching disk I/O on Win2k's Performance Monitor while I issue various "flavors". Our folks use lots of complex query terms with heavy use of the Stemmer. I've gotten them to switch from using tons of ORs to using the Equivalence operator and we're getting much better results using NEAR than simple ANDs. Performance is very good, with CONTAINS queries returning results in less than a second for terms that are rare in the docs, up to a minute for terms that are common in lots (e.g. hundreds of thousands) of docs. If you're going to do synonym searches, you'd better start looking for a good thesaurus - the one Oracle ships is pretty limited. We've not found a good one for the technical lingo our docs contain, so we don't do ABOUT queries at this time. Get familiar with CTX_Query.Explain, it will help you understand things like what the Stemmer *really* does and how complex queries are parsed. Hope this helps. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Kendall Sent: Thursday, October 04, 2001 10:00 AM To: Multiple recipients of list ORACLE-L Hello all, Although I have installed Intermedia as part of my general DBA duties before I have not experienced any particular requirements on throughput rate or indexing. I need some information on being able to deal with large volumes of product data (e.g. 1 million products in a retail application) and be able to perform 'intelligent' searches against the metadata (things like typographical error matching, synonyms etc.) as well as the more usual parametric search (i.e. advanced search page with lots of metadata specific fields). Indexing time and max throughput are also of interest. Any data based on experience would be appreciated. Thanks Martin -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack C. Applewhite INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP
RE: Intermedia Performance Benchmarks anyone ?
Mario, Yes. In 8.1.6.0.0 on Win2k you can't use the SCORE() operator to return the actual document score produced by the CONTAINS operator when the query is executed against a Union View. That prevents us from ordering a result set by the Score() when we query our 6 Month View - a minor annoyance. However, SCORE() works just fine against interMedia Text indexed tables - either "regular" or partitioned. That's the only other real bug I can remember that we've run up against. Jack Jack C. Applewhite Database Administrator/Developer OCP Oracle8 DBA iNetProfit, Inc. Austin, Texas www.iNetProfit.com [EMAIL PROTECTED] (512)327-9068 -Original Message- Sent: Friday, October 05, 2001 11:23 AM To: [EMAIL PROTECTED]; [EMAIL PROTECTED] Excellent doc. I just wonder if there are only cache related bugs on using interMedia. Mario Alberto Ramos -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Jack C. Applewhite INET: [EMAIL PROTECTED] Fat City Network Services-- (858) 538-5051 FAX: (858) 538-5051 San Diego, California-- Public Internet access / Mailing Lists To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).