Re: Non-linear degradation in bulk loads?

2005-10-18 Thread Heikki Tuuri
] Vastaanottaja: Heikki Tuuri [EMAIL PROTECTED] mailto:[EMAIL PROTECTED]; mysql@lists.mysql.com mailto:mysql@lists.mysql.com Lähetetty: Tuesday, October 18, 2005 4:39 AM Aihe: RE: Non-linear degradation in bulk loads? Side question: If I use a 64-bit MySQL build on a 64-bit kernel, is it safe

RE: Non-linear degradation in bulk loads?

2005-10-17 Thread Jon Frisby
About 6GB... Is there any way to forcibly limit this? -JF -Original Message- From: Heikki Tuuri [mailto:[EMAIL PROTECTED] Sent: Sunday, October 16, 2005 2:23 AM To: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? Jon, your hardware/OS combination

RE: Non-linear degradation in bulk loads?

2005-10-17 Thread Jon Frisby
, 2005 10:39 PM Subject: RE: Non-linear degradation in bulk loads? I've tried tweaking the structure of the schema to have, for example, a = PRIMARY KEY index on email, no other indexes, and then insert in sorted = order -- made no improvement whatsoever. Another clue that leads me

Re: Non-linear degradation in bulk loads?

2005-10-17 Thread Heikki Tuuri
: RE: Non-linear degradation in bulk loads? Actually, I believe we're running 32-bit, with bigmem... Does similar behavior occur in such a scenario? -JF -Original Message- From: Heikki Tuuri [mailto:[EMAIL PROTECTED] Sent: Sunday, October 16, 2005 2:23 AM To: mysql@lists.mysql.com

RE: Non-linear degradation in bulk loads?

2005-10-17 Thread Jon Frisby
Sent: Friday, October 14, 2005 10:39 PM Subject: RE: Non-linear degradation in bulk loads? I've tried tweaking the structure of the schema to have, for example, a = PRIMARY KEY index on email, no other indexes, and then insert in sorted = order -- made no improvement whatsoever

Re: Non-linear degradation in bulk loads?

2005-10-17 Thread Heikki Tuuri
: Re: Non-linear degradation in bulk loads? Jon, your hardware/OS combination is quite new and unusual. It might indeed be an OS problem. We observed from a 64-bit RHEL 4 that when the file cache of the OS grew bigger than 4 GB, then the file I/O performance dropped to one tenth of the normal. You

RE: Non-linear degradation in bulk loads?

2005-10-17 Thread Jon Frisby
Message- From: Heikki Tuuri [mailto:[EMAIL PROTECTED] Sent: Monday, October 17, 2005 10:52 AM To: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? Jon, I do not know. Why not install a 64-bit Linux in your computer? Regards, Heikki Oracle/Innobase

RE: Non-linear degradation in bulk loads?

2005-10-17 Thread Jon Frisby
as the sole application on the machine? -JF -Original Message- From: Heikki Tuuri [mailto:[EMAIL PROTECTED] Sent: Monday, October 17, 2005 10:55 AM To: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? Jon, I am not 100 % sure that the problem we saw

Re: Non-linear degradation in bulk loads?

2005-10-16 Thread Heikki Tuuri
[mailto:[EMAIL PROTECTED] Sent: Wednesday, October 12, 2005 8:15 AM To: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? =20 Jon, =20 hmm... maybe one of the indexes inevitably is in a random order. =20 Please post a typical =20 SHOW INNODB STATUS\G =20 when the inserts happen

RE: Non-linear degradation in bulk loads?

2005-10-14 Thread Jon Frisby
] Sent: Wednesday, October 12, 2005 8:15 AM To: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? Jon, hmm... maybe one of the indexes inevitably is in a random order. Please post a typical SHOW INNODB STATUS\G when the inserts happen slowly. What is your

Re: Non-linear degradation in bulk loads?

2005-10-12 Thread Heikki Tuuri
: Heikki Tuuri [EMAIL PROTECTED]; mysql@lists.mysql.com Lähetetty: Wednesday, October 12, 2005 3:08 AM Aihe: RE: Non-linear degradation in bulk loads? Two solutions: 1) sort the rows to be inserted on the key 'email' before inserting. 2) Or: http://dev.mysql.com/doc/mysql/en/innodb

Re: Non-linear degradation in bulk loads?

2005-10-11 Thread Devananda
Jon Frisby wrote: Everyone, We're trying to do some bulk data loads on several different tables (on several different machines, using several different techniques) and seeing dramatically worse-than-linear performance. We've tried the bulk-INSERT syntax, and the LOAD DATA INFILE syntax. We've

Re: Non-linear degradation in bulk loads?

2005-10-11 Thread Heikki Tuuri
Jon, my guess is that the inserts to the UNIQUE secondary index cause the workload to be seriously disk-bound. Two solutions: 1) sort the rows to be inserted on the key 'email' before inserting. 2) Or: http://dev.mysql.com/doc/mysql/en/innodb-tuning.html If you have UNIQUE constraints on

Re: Non-linear degradation in bulk loads?

2005-10-11 Thread mos
At 11:41 PM 10/10/2005, you wrote: Hi Jon, Well, may be the next suggestions might help you. Disable Keys does apply to non-unique keys only. So I suggest to focus on your unique Email key. You could do some tests with: a. drop the unique key on Email b. load the various bulks c. after loading,

RE: Non-linear degradation in bulk loads?

2005-10-11 Thread Jon Frisby
Two solutions: 1) sort the rows to be inserted on the key 'email' before inserting. 2) Or: http://dev.mysql.com/doc/mysql/en/innodb-tuning.html If you have UNIQUE constraints on secondary keys, starting from MySQL 3.23.52 and 4.0.3, you can speed up table imports by temporarily

Re: Non-linear degradation in bulk loads?

2005-10-10 Thread Manoj
Not sure but given that you suffer from non-linear degradation in performance;my guess is you might be extending your ibdata file every too frequently during the batch load process. Check the ibdata_data_file_path variable in my.cnf for more details. Cheers Manoj On 10/11/05, Jon Frisby [EMAIL

RE: Non-linear degradation in bulk loads?

2005-10-10 Thread Jon Frisby
- From: Manoj [mailto:[EMAIL PROTECTED] Sent: Monday, October 10, 2005 8:51 PM To: Jon Frisby Cc: mysql@lists.mysql.com Subject: Re: Non-linear degradation in bulk loads? Not sure but given that you suffer from non-linear degradation in performance;my guess is you might be extending your

Re: Non-linear degradation in bulk loads?

2005-10-10 Thread C.R. Vegelin
Hi Jon, Well, may be the next suggestions might help you. Disable Keys does apply to non-unique keys only. So I suggest to focus on your unique Email key. You could do some tests with: a. drop the unique key on Email b. load the various bulks c. after loading, define Email, eg. as (unique)