Jun 23: Northern Server Technology Day
Last Thursday's Northern Server Technology Day organised by the UKOUG was one of the more enjoyable events I've attended. It helped that there were no hotels, driving or airports involved, just a simple walk to the train station, a 90 minute train journey (free wireless on hand) and a 10 minute walk at the other end. Given such a short trip, I decided to take a lazy approach and didn't turn up until the start of the second presentation of the day - Pete Finnigan talking about Security Tools. Although I'm already familiar with a lot of the tools from Pete's website, it was a nice summary containing a few funny security stories so it was a good start to the day.
Next was Piet de Visser sharing some of his extensive experience with 10g Upgrade Nightmares. I've often said that this is my favourite type of User Group presentation. I'm not mad keen on the term 'war stories' but users telling other users about the problems they've faced (along with the answers) is surely what a user group is for. Piet's two main messages to take away were one that he's mentioned in his previous presentations and one that was upgrade-specific
1) The more simple something is, the better.
2) Try to test against a true file copy of your database and o/s.
So that was two out of two good presentations.
After a quick break for lunch, it was time for mine. The less said about it the better, from a personal point of view. It wasn't that it was terrible (I don't think) and several people were very complimentary about it afterwards, but in an attempt to be informal, I tried to avoid it being too 'scripted'. The fact is that I do have a couple of pages of notes for the presentation, so it wasn't laziness, but chose to ignore them and take a dive around the OEM Performance Pages. It didn't really work, most obviously because I spent the rest of the day remembering things that I would have liked to show people and didn't. So if you didn't know what to expect, it might have come across reasonably, but I know what it should have been like and it was nowhere near. I'm still convinced it's a worthwhile presentation, though, and one I'll revisit. Some of the things I forgot?
- The Trace this session button.
- Any mention of ADDM.
- Event Histograms (although that I don't find that particular screen as useful in 10g, where it only shows the range of values since instance startup.
(Actually, this could go on for a while, so I'll skip to the two most important ...)
- The acknowledgements slide, thanking some of the guys at Oracle for their help with questions.
- Any mention of Kyle Hailey's Simulated ASH or ASH Masters.
Fortunately, I was able to point a few interested parties to the links immediately after the presentation so hopefully ASH Masters will have a few more visitors this week.
It helped me get over my disappointment that Jonathan Lewis' Trouble-shooting presentation was next. I thought it was one of the best 45 minute presentations I've seen him deliver because it dealt with an overall approach to performance analysis that utilised the various tools available and showed how they related to each other. Two of the most important sections of the presentation were near the start.
First, he broke down the Service and Wait components of R= S + W into time spent competing for resources and time spent using resources. For example, from an I/O wait perspective, some wait time is inevitable whenever you use resources (disk access time, or even cache access time) whereas some wait time could be down to queuing, unable to perform the required work. There was a bit more to it than that, but hopefully you get the general idea.
He also described the available statistics from the V$ views (or rather their low-level sources) as being in four dimensions.
Session (or Service or Systems, at higher levels)
SQL Statement
Statistic Type (e.g. Segment Statistics, Latch Statistics and so on)
By adding the fourth dimension of time, we build up a cube of data and how we analyse that cube depends on the nature of the problem we're trying to solve.
It was a terrific presentation in my view because it often isn't the technical details that people don't understand when tuning performance problems, but the overall approach which will suit different types of problem. It's one of the reasons why I wanted to do an introductory presentation on this very subject at UKOUG this year. Now I just need to find a way of saying the same things without saying the same things
At the end Peter Robson, one of the directors of the UKOUG, asked Jonathan a question about complexity and Oracle's RDBMS but I don't think I understood the question fully and I'm not sure Jonathan understood it either or answered it correctly, so I'm trying to encourage Peter to write another guest blog entry. Fingers crossed.
Afterwards I had a few beers with several of the speakers, Lisa Dobson who had helped organise the day with the usual UKOUG support and a few old friends. (Which isn't to say that the first two groups aren't sub-sets of the third) Apologies to Peter Lewis of the BGS for managing to spill my pint over him and his bag! I'm sure whoever was sitting near him on his train home was grateful
So it was an enjoyable and interesting day out of the office and I'm glad I eventually accepted Lisa's invitation. Thanks to everyone at the UKOUG who made it a successful day. I have to close on a negative note though (what a surprise) and the latest of a series of reminders to myself.
I must not talk to my mates during presentations
I must not talk to my mates during presentations
I must not .....
Next was Piet de Visser sharing some of his extensive experience with 10g Upgrade Nightmares. I've often said that this is my favourite type of User Group presentation. I'm not mad keen on the term 'war stories' but users telling other users about the problems they've faced (along with the answers) is surely what a user group is for. Piet's two main messages to take away were one that he's mentioned in his previous presentations and one that was upgrade-specific
1) The more simple something is, the better.
2) Try to test against a true file copy of your database and o/s.
So that was two out of two good presentations.
After a quick break for lunch, it was time for mine. The less said about it the better, from a personal point of view. It wasn't that it was terrible (I don't think) and several people were very complimentary about it afterwards, but in an attempt to be informal, I tried to avoid it being too 'scripted'. The fact is that I do have a couple of pages of notes for the presentation, so it wasn't laziness, but chose to ignore them and take a dive around the OEM Performance Pages. It didn't really work, most obviously because I spent the rest of the day remembering things that I would have liked to show people and didn't. So if you didn't know what to expect, it might have come across reasonably, but I know what it should have been like and it was nowhere near. I'm still convinced it's a worthwhile presentation, though, and one I'll revisit. Some of the things I forgot?
- The Trace this session button.
- Any mention of ADDM.
- Event Histograms (although that I don't find that particular screen as useful in 10g, where it only shows the range of values since instance startup.
(Actually, this could go on for a while, so I'll skip to the two most important ...)
- The acknowledgements slide, thanking some of the guys at Oracle for their help with questions.
- Any mention of Kyle Hailey's Simulated ASH or ASH Masters.
Fortunately, I was able to point a few interested parties to the links immediately after the presentation so hopefully ASH Masters will have a few more visitors this week.
It helped me get over my disappointment that Jonathan Lewis' Trouble-shooting presentation was next. I thought it was one of the best 45 minute presentations I've seen him deliver because it dealt with an overall approach to performance analysis that utilised the various tools available and showed how they related to each other. Two of the most important sections of the presentation were near the start.
First, he broke down the Service and Wait components of R= S + W into time spent competing for resources and time spent using resources. For example, from an I/O wait perspective, some wait time is inevitable whenever you use resources (disk access time, or even cache access time) whereas some wait time could be down to queuing, unable to perform the required work. There was a bit more to it than that, but hopefully you get the general idea.
He also described the available statistics from the V$ views (or rather their low-level sources) as being in four dimensions.
Session (or Service or Systems, at higher levels)
SQL Statement
Statistic Type (e.g. Segment Statistics, Latch Statistics and so on)
By adding the fourth dimension of time, we build up a cube of data and how we analyse that cube depends on the nature of the problem we're trying to solve.
It was a terrific presentation in my view because it often isn't the technical details that people don't understand when tuning performance problems, but the overall approach which will suit different types of problem. It's one of the reasons why I wanted to do an introductory presentation on this very subject at UKOUG this year. Now I just need to find a way of saying the same things without saying the same things
At the end Peter Robson, one of the directors of the UKOUG, asked Jonathan a question about complexity and Oracle's RDBMS but I don't think I understood the question fully and I'm not sure Jonathan understood it either or answered it correctly, so I'm trying to encourage Peter to write another guest blog entry. Fingers crossed.
Afterwards I had a few beers with several of the speakers, Lisa Dobson who had helped organise the day with the usual UKOUG support and a few old friends. (Which isn't to say that the first two groups aren't sub-sets of the third
) Apologies to Peter Lewis of the BGS for managing to spill my pint over him and his bag! I'm sure whoever was sitting near him on his train home was grateful So it was an enjoyable and interesting day out of the office and I'm glad I eventually accepted Lisa's invitation. Thanks to everyone at the UKOUG who made it a successful day. I have to close on a negative note though (what a surprise) and the latest of a series of reminders to myself.
I must not talk to my mates during presentations
I must not talk to my mates during presentations
I must not .....
Jun 22: Oracle Security Survey
Jun 10: More Blogs
It's good to see some more people joining in, particularly those I already know and admire.
First up is Mark Bobak, who I've met at several conferences and, as you can tell from his first few posts, is likely to have some interesting technical tales to tell.
Via Mark's blog-roll, I noticed Robyn Sands has been blogging for a while, too. As well as having shared DBA interests (being oddly organised; OFA and anything else that brings some engineering discipline to our field), we have the same taste in presentation images.
Next is Karen Morton of Method R Corp., Cary Millsap's new company, who has an interesting blog post about presentation style. Maybe I'll comment more on that later, but it's been a very long (if interesting) day at work and Ibrahimovic just scored the first goal after I missed laying the bet on him that I'd planned
Oh, and to revisit an old favourite - do you think that the Dutch are quite pleased about last night's 3-0 victory against Italy? (I don't know whether he's changed the oranje, but it seems much calmer than how it looked in the office today!) They have every right to be too, but don't tell me that prone Italian defender was playing Van Nistelrooy onside! (Ooops, slipped into a personal blog moment there)
First up is Mark Bobak, who I've met at several conferences and, as you can tell from his first few posts, is likely to have some interesting technical tales to tell.
Via Mark's blog-roll, I noticed Robyn Sands has been blogging for a while, too. As well as having shared DBA interests (being oddly organised; OFA and anything else that brings some engineering discipline to our field), we have the same taste in presentation images.
Next is Karen Morton of Method R Corp., Cary Millsap's new company, who has an interesting blog post about presentation style. Maybe I'll comment more on that later, but it's been a very long (if interesting) day at work and Ibrahimovic just scored the first goal after I missed laying the bet on him that I'd planned
Oh, and to revisit an old favourite - do you think that the Dutch are quite pleased about last night's 3-0 victory against Italy? (I don't know whether he's changed the oranje, but it seems much calmer than how it looked in the office today!) They have every right to be too, but don't tell me that prone Italian defender was playing Van Nistelrooy onside! (Ooops, slipped into a personal blog moment there)
Jun 8: Advert - Future Plans
After over a year at my current site, I've decided that it's time for something new. I'm currently involved in a Grid Control and Audit Vault implementation project but I'll only be spending another 6 weeks on that. My current client have one of their best permies working on it, too, and 7 weeks gives them plenty of time to look for a replacement contractor if they decide that's what they want.
So, what next? Well I want some time to play and I don't mean Wii games. Maybe I could say I want to do some research but that seems a little pretentious for my usual efforts. Let me just say that I want to spend a month working from my office at home doing nothing but looking at Oracle, investigating a few features and probably reading some of those books in my massive pile - refreshing my technical skills, in other words. Maybe I'll write a paper or two.
I do know that I've been putting off some improvements to the 10g Performance Course before I teach it again. Whilst the first public teach in Prague went very well, judging by the feedback from the attendees and Oracle, I could think of several things I wanted to improve. My highest priorities are improving the agenda and including more detail in the course notes and that will take more time than is left spare when I'm on a client site five days a week and have another life away from Oracle. There's a possibility I'll be teaching it in Austria in August, so keep an eye out for that, and I have reasonably firm dates of 16th and 17th October in Belgium (Kurt insisted I should mention it soon).
However, I'd also like to teach an additional course at home during my non-client time, so I'm keen to determine if there's any interest in a course in the UK during August. I think it would be a fun event if I could run a small course in Edinburgh while the Festival is on. Accommodation is hard to come by at that time of year (let me worry about that, though), but a couple of days in Scotland with some decent learning and discussion during the day, followed by pleasant sunny evenings when Edinburgh is at it's best seems an attractive proposition to me. Then again, a course in London might be easier for people to attend? I'm open to ideas, but if anyone is interested in the possibility, then leave a comment here or drop me an email at my usual email address (you can find that on my cv elsewhere on this site if you don't know it already). Because I want to teach more again, I'm not looking to make a big profit from the event. However, I expect a small number of attendees without the marketing might of a large training company (which is good, from the attendees point of view) so I'm looking at a guide price of £500-600 for the two days. Accommodation and transport would be at your own expense.
Once I've updated the agenda, I'll post it here in the next couple of days.
Oh, but here's the emotional clincher My 16 year old daughter has an interest in event management so, if we do go ahead, it'll give us a chance to work together on putting on a two day event we can both be proud of, before she goes off to college.
So, what next? Well I want some time to play and I don't mean Wii games. Maybe I could say I want to do some research but that seems a little pretentious for my usual efforts. Let me just say that I want to spend a month working from my office at home doing nothing but looking at Oracle, investigating a few features and probably reading some of those books in my massive pile - refreshing my technical skills, in other words. Maybe I'll write a paper or two.
I do know that I've been putting off some improvements to the 10g Performance Course before I teach it again. Whilst the first public teach in Prague went very well, judging by the feedback from the attendees and Oracle, I could think of several things I wanted to improve. My highest priorities are improving the agenda and including more detail in the course notes and that will take more time than is left spare when I'm on a client site five days a week and have another life away from Oracle. There's a possibility I'll be teaching it in Austria in August, so keep an eye out for that, and I have reasonably firm dates of 16th and 17th October in Belgium (Kurt insisted I should mention it soon).
However, I'd also like to teach an additional course at home during my non-client time, so I'm keen to determine if there's any interest in a course in the UK during August. I think it would be a fun event if I could run a small course in Edinburgh while the Festival is on. Accommodation is hard to come by at that time of year (let me worry about that, though), but a couple of days in Scotland with some decent learning and discussion during the day, followed by pleasant sunny evenings when Edinburgh is at it's best seems an attractive proposition to me. Then again, a course in London might be easier for people to attend? I'm open to ideas, but if anyone is interested in the possibility, then leave a comment here or drop me an email at my usual email address (you can find that on my cv elsewhere on this site if you don't know it already). Because I want to teach more again, I'm not looking to make a big profit from the event. However, I expect a small number of attendees without the marketing might of a large training company (which is good, from the attendees point of view) so I'm looking at a guide price of £500-600 for the two days. Accommodation and transport would be at your own expense.
Once I've updated the agenda, I'll post it here in the next couple of days.
Oh, but here's the emotional clincher
My 16 year old daughter has an interest in event management so, if we do go ahead, it'll give us a chance to work together on putting on a two day event we can both be proud of, before she goes off to college. Jun 8: Real Application Testing and more Relinking
I'd been thinking of blogging about the appearance of Real Application Testing in 10.2.0.4 since I noticed it post-upgrade, just before the course in Prague.
It looked like a nice addition because, as others have pointed out on their blogs, it could prove very useful for capturing an application workload on 10g to test how it will work on 11g before upgrading. There's a genuine business need for that. Then, having updated our first couple of test instances at work, I was reminded of the great care we take with license management at my current site (thanks, Tam) and, on closer inspection of the documentation :-
Whilst I realise that's a link to the 11g documentation, I don't imagine for a second that Oracle won't apply the same license rules to the 10.2.0.4 or 9.2.0.8 instances where the workload is captured. The additional cost is around £5000 per processor.
So it looked like it was time for another relink operation. (Note that this operation was performed on an AIX server. Windows is very different!) It didn't take long to have a look in ins_rdbms.mk and search for 'rat'. Sure enough, there are options to enable or disable RAT. So our new standard relink command included in our installation process is.
One or two people have questioned our relinking of the product because it means we're running on an executable that few sites are using and I have some sympathy with that view, but it's far too easy for a DBA to think 'Mmmm, I'll try that out' or not even be aware that they're trying out an additional cost option. Oh, and if you think Oracle wouldn't know about it, then how about this view?
I haven't used workload capture on this instance yet, so I thought I'd give it a quick try. Interesting though it is to try this out for the first time, I'm going to keep this post short by referring you to another blog that shows how to use RAT via OEM DB Console. Actually, my first attempt failed with the following error message.
OK, so let me fix that and try again
Yes, it worked this time. Here's the Summary Screen and you can pick up the associated capture report from here. Note that this was a 5 minute/2 user test of the SOE benchmark that's supplied with Swingbench.
More to the point, maybe the feature usage view has been updated?
Perhaps not, but I expect it will be next time this sample is updated.
Connected to:
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options
It looked like a nice addition because, as others have pointed out on their blogs, it could prove very useful for capturing an application workload on 10g to test how it will work on 11g before upgrading. There's a genuine business need for that. Then, having updated our first couple of test instances at work, I was reminded of the great care we take with license management at my current site (thanks, Tam) and, on closer inspection of the documentation :-
The Oracle Real Application Testing license is required on both capture and replay systems for Database Replay and is charged by the total number of CPUs on those systems.
Whilst I realise that's a link to the 11g documentation, I don't imagine for a second that Oracle won't apply the same license rules to the 10.2.0.4 or 9.2.0.8 instances where the workload is captured. The additional cost is around £5000 per processor.
So it looked like it was time for another relink operation. (Note that this operation was performed on an AIX server. Windows is very different!) It didn't take long to have a look in ins_rdbms.mk and search for 'rat'. Sure enough, there are options to enable or disable RAT. So our new standard relink command included in our installation process is.
/usr/ccs/bin/make -f ins_rdbms.mk olap_off dm_off rat_off ioracle
One or two people have questioned our relinking of the product because it means we're running on an executable that few sites are using and I have some sympathy with that view, but it's far too easy for a DBA to think 'Mmmm, I'll try that out' or not even be aware that they're trying out an additional cost option. Oh, and if you think Oracle wouldn't know about it, then how about this view?
TESTUSER@TEST1020> select name, detected_usages, currently_used,
2 first_usage_date, feature_info, last_sample_date,
3 sample_interval, description
4 from DBA_FEATURE_USAGE_STATISTICS
5 where name like 'Database Replay%'
6 /
NAME DETECTED_USAGES CURRE FIRST_USA
---------------------------------------- --------------- ----- ---------
FEATURE_INFO
--------------------------------------------------------------------------------
LAST_SAMP SAMPLE_INTERVAL
--------- ---------------
DESCRIPTION
--------------------------------------------------------------------------------
Database Replay: Workload Capture 0 FALSE
20-MAY-08 604800
Database Replay: Workload was ever captured.
I haven't used workload capture on this instance yet, so I thought I'd give it a quick try. Interesting though it is to try this out for the first time, I'm going to keep this post short by referring you to another blog that shows how to use RAT via OEM DB Console. Actually, my first attempt failed with the following error message.
SYSTEM@TEST1020> BEGIN dbms_workload_capture.start_capture('CAPTURE-TEST1020-20080608151755',
'DATA_FILE_DIR', 300, 'INCLUDE', FALSE); END;
*
ERROR at line 1:
ORA-15591: cannot start capture because parameter "PRE_11G_ENABLE_CAPTURE" is
not enabled
ORA-06512: at "SYS.DBMS_WORKLOAD_CAPTURE", line 799
ORA-06512: at line 1OK, so let me fix that and try again
SYS@TEST1020> alter system set pre_11g_enable_capture=true;
System altered.
Yes, it worked this time. Here's the Summary Screen and you can pick up the associated capture report from here. Note that this was a 5 minute/2 user test of the SOE benchmark that's supplied with Swingbench.
More to the point, maybe the feature usage view has been updated?
SYS@TEST1020> select name, detected_usages, currently_used,
2 first_usage_date, feature_info, last_sample_date,
3 sample_interval, description
4 from DBA_FEATURE_USAGE_STATISTICS
5 where name like 'Database Replay%';
NAME DETECTED_USAGES CURRE FIRST_USA
---------------------------------------- --------------- ----- ---------
FEATURE_INFO
--------------------------------------------------------------------------------
LAST_SAMP SAMPLE_INTERVAL
--------- ---------------
DESCRIPTION
--------------------------------------------------------------------------------
Database Replay: Workload Capture 0 FALSE
08-JUN-08 604800
Database Replay: Workload was ever captured.
Perhaps not, but I expect it will be next time this sample is updated.
Comments
Thu, 03.07.2008 11:08
As a continuation, I raised a SR with Oracle on this and had confirmation that [...]
Fri, 27.06.2008 09:03
Thanks Doug, nice to see someone picked up the messages. It was a good [...]
Tue, 24.06.2008 14:00
Yes, it was over all a very good day thanks to Lisa and the presenters. Doug [...]