Re: SharePlex info
Title: Message Sorry about the late reply but (if I remember correctly from my research about one year ago) Shareplex does something like log mining only on Unix systems. On NT it uses triggers just like replication. Yechiel AdarMehish - Original Message - From: Aponte, Tony To: Multiple recipients of list ORACLE-L Sent: Friday, August 22, 2003 10:39 PM Subject: SharePlex info You are correct in the first place. SharePlex works as you describe, it mines the log and sends only the absolute minimum to reassemble the transaction on the target. It doesn't send SQL. The target side processes take the data and rebuild a SQL statement from the DDL definitions it got from the data dictionaries of the source and target (just in case you only want a subset of the columns.) Sorry if I confused you. Tony -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:01 PMTo: Aponte, TonySubject: RE: SharePlex info Tony, My question was inspired by belief that SharePlex does log mining on the source DB and hence do not send unnecessary data over the network. Apparently, this is not the case. I didn't want to compare SharePlex to logical standby cause I know that logical standby definitely needs all logs transported to the target site where is does log mining. We considering remote disaster recovery site where we want to have working data and we don't care much about "log" tables. Thank you for valuable info. -Original Message-From: Aponte, Tony [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 4:40 PMTo: [EMAIL PROTECTED]Cc: [EMAIL PROTECTED]Subject: RE: SharePlex info Your bandwidth requirements will be the rate of changes to the actual data. The traffic consists of the actual data and control information needed to reassemble the transaction on the target. The source database's other redo payload (i.e., index operations, rollback segment maintenance, etc.) is not used by Shareplex. In our environment of dual Sun 6800's, 10 CPU's each, we observe less that 1% CPU consumption on the source and target sides combined. It varies according to the DML load on the source but not by much. We've never had a problem with it consuming a noticeable amount. I have a question on the comparison between a physical standby and Shareplex replication.Isn't9i's logical standby featurebetter suited for the comparison to Shareplex? I'm assuming that you are considering offloading some processing to another host since you are looking to replicate about 50% of the tables in the source database. HTH Tony Aponte -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 1:49 PMTo: Multiple recipients of list ORACLE-LSubject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
RE: SharePlex info
Title: Message Shareplex does not use triggers on NT it uses the same underlying technology as it does on Unix "reading" the log files and shipping SQL to the target database. It uses a 3rd party tool called "Knutcracker" to allow it to some ofits UNIX commands on NT. T¬-Original Message-From: Yechiel Adar [mailto:[EMAIL PROTECTED]Sent: 25 August 2003 09:10To: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info Sorry about the late reply but (if I remember correctly from my research about one year ago) Shareplex does something like log mining only on Unix systems. On NT it uses triggers just like replication. Yechiel AdarMehish - Original Message - From: Aponte, Tony To: Multiple recipients of list ORACLE-L Sent: Friday, August 22, 2003 10:39 PM Subject: SharePlex info You are correct in the first place. SharePlex works as you describe, it mines the log and sends only the absolute minimum to reassemble the transaction on the target. It doesn't send SQL. The target side processes take the data and rebuild a SQL statement from the DDL definitions it got from the data dictionaries of the source and target (just in case you only want a subset of the columns.) Sorry if I confused you. Tony -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:01 PMTo: Aponte, TonySubject: RE: SharePlex info Tony, My question was inspired by belief that SharePlex does log mining on the source DB and hence do not send unnecessary data over the network. Apparently, this is not the case. I didn't want to compare SharePlex to logical standby cause I know that logical standby definitely needs all logs transported to the target site where is does log mining. We considering remote disaster recovery site where we want to have working data and we don't care much about "log" tables. Thank you for valuable info. -Original Message-From: Aponte, Tony [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 4:40 PMTo: [EMAIL PROTECTED]Cc: [EMAIL PROTECTED]Subject: RE: SharePlex info Your bandwidth requirements will be the rate of changes to the actual data. The traffic consists of the actual data and control information needed to reassemble the transaction on the target. The source database's other redo payload (i.e., index operations, rollback segment maintenance, etc.) is not used by Shareplex. In our environment of dual Sun 6800's, 10 CPU's each, we observe less that 1% CPU consumption on the source and target sides combined. It varies according to the DML load on the source but not by much. We've never had a problem with it consuming a noticeable amount. I have a question on the comparison between a physical standby and Shareplex replication.Isn't9i's logical standby featurebetter suited for the comparison to Shareplex? I'm assuming that you are considering offloading some processing to another host since you are looking to replicate about 50% of the tables in the source database. HTH Tony Aponte -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 1:49 PMTo: Multiple recipients of list ORACLE-LSubject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
RE: SharePlex info
Title: Message Tony, My question was inspired by belief that SharePlex does log mining on the source DB and hence do not send unnecessary data over the network. Apparently, this is not the case. I didn't want to compare SharePlex to logical standby cause I know that logical standby definitely needs all logs transported to the target site where is does log mining. We considering remote disaster recovery site where we want to have working data and we don't care much about "log" tables. Thank you for valuable info. -Original Message-From: Aponte, Tony [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:44 PMTo: Multiple recipients of list ORACLE-LSubject: RE: SharePlex info Your bandwidth requirements will be the rate of changes to the actual data. The traffic consists of the actual data and control information needed to reassemble the transaction on the target. The source database's other redo payload (i.e., index operations, rollback segment maintenance, etc.) is not used by Shareplex. In our environment of dual Sun 6800's, 10 CPU's each, we observe less that 1% CPU consumption on the source and target sides combined. It varies according to the DML load on the source but not by much. We've never had a problem with it consuming a noticeable amount. I have a question on the comparison between a physical standby and Shareplex replication.Isn't9i's logical standby featurebetter suited for the comparison to Shareplex? I'm assuming that you are considering offloading some processing to another host since you are looking to replicate about 50% of the tables in the source database. HTH Tony Aponte -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 1:49 PMTo: Multiple recipients of list ORACLE-LSubject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
Re: SharePlex info
1)You would need lessnetwork bandwidth with shareplex than you would for transporting archive logs. about 1/3 rd ofwhat you would need for physical stdby. 2) CPU burden would be 'little' I guess. 3) Shareplex replication allows you to have the table available for read on the target. (even update). If you need this or if it is a great advantage then you can consider shareplex. Else physical stdby would be better. You have to basically consider the huge cost of shareplex and maintenance it needs.CPU usageof source would be a lesser consideration i think. somedaysback there was a thread on this started by oneNelson ( I think)so you can maybe look at the archives and thatshould help you make a decision. "Gorbounov,Vadim" [EMAIL PROTECTED] wrote: Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim Do you Yahoo!? SBC Yahoo! DSL - Now only $29.95 per month!
Re: SharePlex info
Title: Message Hi! Btw, you can physically replicate 50% of your tables with regular standby mechanisms as well. You just take the files belonging to non-needed tablespaces offline and standby recovers only the required part. You just have to arrange your tables to right tablespaces and spend your money elsewhere. Physical standby and shareplex can operate on archivelogs, thus they can do their jobs without any additional burden to source database CPU, since you generate and archive your logs anyway. You can do archivelog's processing on target or some staging server. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Thursday, August 21, 2003 8:49 PM Subject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
RE: SharePlex info
Thank you, Raju. Very helpful -Original Message-From: raju pa [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 4:59 PMTo: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info 1)You would need lessnetwork bandwidth with shareplex than you would for transporting archive logs. about 1/3 rd ofwhat you would need for physical stdby. 2) CPU burden would be 'little' I guess. 3) Shareplex replication allows you to have the table available for read on the target. (even update). If you need this or if it is a great advantage then you can consider shareplex. Else physical stdby would be better. You have to basically consider the huge cost of shareplex and maintenance it needs.CPU usageof source would be a lesser consideration i think. somedaysback there was a thread on this started by oneNelson ( I think)so you can maybe look at the archives and thatshould help you make a decision. "Gorbounov,Vadim" [EMAIL PROTECTED] wrote: Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim Do you Yahoo!?SBC Yahoo! DSL - Now only $29.95 per month!
RE: SharePlex info
Title: Message Your bandwidth requirements will be the rate of changes to the actual data. The traffic consists of the actual data and control information needed to reassemble the transaction on the target. The source database's other redo payload (i.e., index operations, rollback segment maintenance, etc.) is not used by Shareplex. In our environment of dual Sun 6800's, 10 CPU's each, we observe less that 1% CPU consumption on the source and target sides combined. It varies according to the DML load on the source but not by much. We've never had a problem with it consuming a noticeable amount. I have a question on the comparison between a physical standby and Shareplex replication.Isn't9i's logical standby featurebetter suited for the comparison to Shareplex? I'm assuming that you are considering offloading some processing to another host since you are looking to replicate about 50% of the tables in the source database. HTH Tony Aponte -Original Message-From: Gorbounov,Vadim [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 1:49 PMTo: Multiple recipients of list ORACLE-LSubject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
RE: SharePlex info
Title: Message Tanel, That's nice trick, thanks a lot. In this casewhole redo steam must be passed over the network anyway. 5 MB/sec over WAN. So we'are doing research if we could same some bandwidth. Vadim -Original Message-From: Tanel Poder [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:14 PMTo: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info Hi! Btw, you can physically replicate 50% of your tables with regular standby mechanisms as well. You just take the files belonging to non-needed tablespaces offline and standby recovers only the required part. You just have to arrange your tables to right tablespaces and spend your money elsewhere. Physical standby and shareplex can operate on archivelogs, thus they can do their jobs without any additional burden to source database CPU, since you generate and archive your logs anyway. You can do archivelog's processing on target or some staging server. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Thursday, August 21, 2003 8:49 PM Subject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
Re: SharePlex info
But you would be wanting to transfer *full* logfiles away from your production servers anyway at least if your data is worth something... Tanel. 1)You would need lessnetwork bandwidth with shareplex than you would for transporting archive logs. about 1/3 rd ofwhat you would need for physical stdby.
Re: SharePlex info
Title: Message Ok, in this case Shareplex might be better, if it is able to extract only relevant data from logs. Actually, you could dosomewhat similar yourself using logminer as well. You just extract all needed DML statements on either production or staging server, compress the output (because file with sql statements is going to be big but will compress well) and send over network, or go with Oracle Streams rightaway.There's going to be an issue with some special datatypes though IIRC. I don't know the pricing of shareplex, maybe it'd be easier going with it.. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Friday, August 22, 2003 12:54 AM Subject: RE: SharePlex info Tanel, That's nice trick, thanks a lot. In this casewhole redo steam must be passed over the network anyway. 5 MB/sec over WAN. So we'are doing research if we could same some bandwidth. Vadim -Original Message-From: Tanel Poder [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:14 PMTo: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info Hi! Btw, you can physically replicate 50% of your tables with regular standby mechanisms as well. You just take the files belonging to non-needed tablespaces offline and standby recovers only the required part. You just have to arrange your tables to right tablespaces and spend your money elsewhere. Physical standby and shareplex can operate on archivelogs, thus they can do their jobs without any additional burden to source database CPU, since you generate and archive your logs anyway. You can do archivelog's processing on target or some staging server. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Thursday, August 21, 2003 8:49 PM Subject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim
RE: SharePlex info
Yes. A nice neat trick indeed. Has anyone tried this? About your redo generation : 5MB/sec - 18000 MB/hour == 18GB IT is indeed huge. IS this peak or average? Good luck. "Gorbounov,Vadim" [EMAIL PROTECTED] wrote: Tanel, That's nice trick, thanks a lot. In this casewhole redo steam must be passed over the network anyway. 5 MB/sec over WAN. So we'are doing research if we could same some bandwidth. Vadim -Original Message-From: Tanel Poder [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:14 PMTo: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info Hi! Btw, you can physically replicate 50% of your tables with regular standby mechanisms as well. You just take the files belonging to non-needed tablespaces offline and standby recovers only the required part. You just have to arrange your tables to right tablespaces and spend your money elsewhere. Physical standby and shareplex can operate on archivelogs, thus they can do their jobs without any additional burden to source database CPU, since you generate and archive your logs anyway. You can do archivelog's processing on target or some staging server. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Thursday, August 21, 2003 8:49 PM Subject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim Do you Yahoo!? Yahoo! SiteBuilder - Free, easy-to-use web site design software
Re: SharePlex info
It's documented in 8.1.7 docs: http://download-west.oracle.com/docs/cd/A87860_01/doc/server.817/a76995/standbym.htm#27264 In 9.2 docs I didn't find it with brief search... Tanel. - Original Message - From: A Joshi To: Multiple recipients of list ORACLE-L Sent: Friday, August 22, 2003 2:14 AM Subject: RE: SharePlex info Yes. A nice neat trick indeed. Has anyone tried this? About your redo generation : 5MB/sec - 18000 MB/hour == 18GB IT is indeed huge. IS this peak or average? Good luck. "Gorbounov,Vadim" [EMAIL PROTECTED] wrote: Tanel, That's nice trick, thanks a lot. In this casewhole redo steam must be passed over the network anyway. 5 MB/sec over WAN. So we'are doing research if we could same some bandwidth. Vadim -Original Message-From: Tanel Poder [mailto:[EMAIL PROTECTED]Sent: Thursday, August 21, 2003 5:14 PMTo: Multiple recipients of list ORACLE-LSubject: Re: SharePlex info Hi! Btw, you can physically replicate 50% of your tables with regular standby mechanisms as well. You just take the files belonging to non-needed tablespaces offline and standby recovers only the required part. You just have to arrange your tables to right tablespaces and spend your money elsewhere. Physical standby and shareplex can operate on archivelogs, thus they can do their jobs without any additional burden to source database CPU, since you generate and archive your logs anyway. You can do archivelog's processing on target or some staging server. Tanel. - Original Message - From: Gorbounov,Vadim To: Multiple recipients of list ORACLE-L Sent: Thursday, August 21, 2003 8:49 PM Subject: SharePlex info Hi All, I'm trying to find some technical details about SharePlex, that is: - How much network bandwidth I'd expect to replicate from database, generating 1-5 MB/sec redo. DoesSharePlex send SQL text over the network or data in some internal (hopefully compressed)format - How much CPU on the source DB server side would it cost - just a ball park - very little- little - or a lot - Of two options, using 9.2 physical async standby db and clone whole database vs replicate 50% (enough from business requirements) of tables using SharePlex, which onesounds preferrable keeping in mind minimizing CPU burden on the source database. Any opinion or pointer to any benchmark is highly appreciated. Thanks a lot Vadim Do you Yahoo!?Yahoo! SiteBuilder - Free, easy-to-use web site design software