Re: Quesiton about best way to index attributes
On Thu, Mar 1, 2012 at 11:44 AM, Emmanuel Lécharny elecha...@gmail.com wrote: Le 2/29/12 7:50 PM, Kiran Ayyagari a écrit : On Wed, Feb 29, 2012 at 11:44 PM,carlo.acco...@ibs-ag.com wrote: Hi, We have apacheds 2.0-M5 db setup with all ~90k users under one OU. Yes, I know they should be stored with some hierarchy but that's not an option for this particular customer. Having 90K users in one single OU is *not* a bad practice. You may want to setup a hierarchy, but this is not mandatory. I want to optimize searching by lastname, (sn), firstname (givenName) and displayName. I have created indexes under the partition and we see ads-indexcachesize is set by default to 100. Assuming the value correlates to a value for the attribute, Can I set this to 50,000 for each of the attribs?. that won't help much, setting up more memory for the JVM will help You still can increase the number to a higher value, but you must know that as soon as the higher BTree pages will be loaded, the speedup will be lower. With 100 pages loaded, you have most of the BTree loaded. If you have enough memory, just increase this value to 50 000, should not harm. In any case, I would suggest that once you have increased those values for each index that you do a full search to load all of them in memory, to check that you don't get an OOM, before running in production. Also, is there a way to rebuild the indexes in 2.0? I can't seem to find how that's done. no, one way is to delete the index and restart the server followed by adding the index and restart, then the server will automatically build the index during startup Kiran : haven't we added some utility tool to do that ? (It's a bit far in the past.. If we don't have them, we must add them) yes, I have added but was not maintained after that I'm wondering if we don't have a JIRA for that... AFAIR there is one for generic tooling -- Regards, Cordialement, Emmanuel Lécharny www.iktek.com -- Kiran Ayyagari
Quesiton about best way to index attributes
Hi, We have apacheds 2.0-M5 db setup with all ~90k users under one OU. Yes, I know they should be stored with some hierarchy but that's not an option for this particular customer. I want to optimize searching by lastname, (sn), firstname (givenName) and displayName. I have created indexes under the partition and we see ads-indexcachesize is set by default to 100. Assuming the value correlates to a value for the attribute, Can I set this to 50,000 for each of the attribs?. I can dedicate ~1GB of heap space for the ApacheDS JVM. Also, is there a way to rebuild the indexes in 2.0? I can't seem to find how that's done. For example In the UI someone might enter Smit looking for Smith. This gets converted into the following query with the OU as the baseDN. ( (objectclass=inetOrgPerson) (| (displayname=Smit*) (sn=Smit*) (givenName=Smit*) ) ) Lastly, would we be better off performing three separate queries (for displayname,sn,givenName) or will the logical OR search above break the moment it finds a match? Thanks.
Re: Quesiton about best way to index attributes
On Wed, Feb 29, 2012 at 11:44 PM, carlo.acco...@ibs-ag.com wrote: Hi, We have apacheds 2.0-M5 db setup with all ~90k users under one OU. Yes, I know they should be stored with some hierarchy but that's not an option for this particular customer. I want to optimize searching by lastname, (sn), firstname (givenName) and displayName. I have created indexes under the partition and we see ads-indexcachesize is set by default to 100. Assuming the value correlates to a value for the attribute, Can I set this to 50,000 for each of the attribs?. that won't help much, setting up more memory for the JVM will help I can dedicate ~1GB of heap space for the ApacheDS JVM. cool, then set the -Xms and -Xmx flags in the java options Also, is there a way to rebuild the indexes in 2.0? I can't seem to find how that's done. no, one way is to delete the index and restart the server followed by adding the index and restart, then the server will automatically build the index during startup For example In the UI someone might enter Smit looking for Smith. This gets converted into the following query with the OU as the baseDN. ( (objectclass=inetOrgPerson) (| (displayname=Smit*) (sn=Smit*) (givenName=Smit*) ) ) Lastly, would we be better off performing three separate queries (for displayname,sn,givenName) or will the logical OR search above break the moment it finds a match? Thanks. the logical OR breaks as soon as it finds a match -- Kiran Ayyagari
Re: Quesiton about best way to index attributes
Le 2/29/12 7:50 PM, Kiran Ayyagari a écrit : On Wed, Feb 29, 2012 at 11:44 PM,carlo.acco...@ibs-ag.com wrote: Hi, We have apacheds 2.0-M5 db setup with all ~90k users under one OU. Yes, I know they should be stored with some hierarchy but that's not an option for this particular customer. Having 90K users in one single OU is *not* a bad practice. You may want to setup a hierarchy, but this is not mandatory. I want to optimize searching by lastname, (sn), firstname (givenName) and displayName. I have created indexes under the partition and we see ads-indexcachesize is set by default to 100. Assuming the value correlates to a value for the attribute, Can I set this to 50,000 for each of the attribs?. that won't help much, setting up more memory for the JVM will help You still can increase the number to a higher value, but you must know that as soon as the higher BTree pages will be loaded, the speedup will be lower. With 100 pages loaded, you have most of the BTree loaded. If you have enough memory, just increase this value to 50 000, should not harm. In any case, I would suggest that once you have increased those values for each index that you do a full search to load all of them in memory, to check that you don't get an OOM, before running in production. Also, is there a way to rebuild the indexes in 2.0? I can't seem to find how that's done. no, one way is to delete the index and restart the server followed by adding the index and restart, then the server will automatically build the index during startup Kiran : haven't we added some utility tool to do that ? (It's a bit far in the past.. If we don't have them, we must add them) I'm wondering if we don't have a JIRA for that... -- Regards, Cordialement, Emmanuel Lécharny www.iktek.com
