Search

Top 60 Oracle Blogs

Recent comments

RAC performance myths

RAC day with Ohio Oracle User Group

I will be talking about RAC and performance in-depth, with lots of demos, in a RAC day training with Ohio Oracle User group on Nov 16,2015 Monday. Venue for the presentation is Dublin, Ohio.

Agenda for the day:

08:00a – 09:00: Registration / Breakfast

09:00a – 09:15: Announcements -Introduction of the speaker

09:15a – 10:30: Underpinning for Oracle RAC and Clusterware

10:30a – 10:45: Break

10:45a – 11:45: RAC cache fusion internals

11:45a – 01:00: Lunch

01:00p – 02:00: RAC Performance tuning Part 1 – Wait events and object tuning

02:00p – 02:15: Break

02:15p – 03:30: RAC performance tuning Part 2 – locks, library cache locks etc.

03:30p – 03:45: Member Announcements, Gift Drawings

Please RSVP to the co-ordinators so that you will have a seat </p />
</p></div>

    	  	<div class=

Golden rules of RAC performance diagnostics

After collaborating with many performance engineers in a RAC database, I have come to realize that there are common pattern among the (mis)diagnosis. This blog about discussing those issues. I talked about this in Hotsos 2014 conference also.

Golden rules

Here are the golden rules of RAC performance diagnostics. These rules may not apply general RAC configuration issues though.

  1. Beware of top event tunnel vision
  2. Eliminate infrastructure as an issue
  3. Identify problem-inducing instance
  4. Review send-side metrics also
  5. Use histograms, not just averages

Looks like, this may be better read as a document. So, please use the pdf files of the presentation and a paper. Presentation slide #10 shows indepth coverage on gc buffer busy* wait events. I will try to blog about that slide later (hopefully).

_gc_fusion_compression

We know that database blocks are transferred between the nodes through the interconnect, aka cache fusion traffic. Common misconception is that packet transfer size is always database block size for block transfer (Of course, messages are smaller in size). That’s not entirely true. There is an optimization in the cache fusion code to reduce the packet size (and so reduces the bits transferred over the private network). Don’t confuse this note with Jumbo frames and MTU size, this note is independent of MTU setting.

RMOUG 2010: My presentations

It is very disappointing to me that I had to cancel my trip to RMOUG training days. I am sick and was not able to catch the flight due to that.

But, I can always share my presentations here. I had two presentations planned in this training day and can be accessed as below:

Advanced RAC troubleshooting
Riyaj_Advanced_rac_troubleshooting_RMOUG_2010_doc
Riyaj_Advanced_rac_troubleshooting_RMOUG_2010_ppt

Why optimizer hates my sql
Riyaj_Why_optimizer_hates_my_sql_2010

RMOUG training days audience: Please accept my sincere apologies.

RAC performance tuning: Understanding Global cache performance

Global cache performance metrics are not correctly measured. It is not understood clearly either. There are even few blogs and web pages disseminating incorrect information. This blog entry is an attempt to offer few methods and scripts to understand global cache performance.

Always review all instances

It is very important to review the performance metrics from all instances in that RAC cluster, not just one instance that you are connected. If you have access to AWR reports, then it is critical to generate AWR reports (or statspack reports) from all instances. But, the problem is that, DBAs tend to generate AWR reports after logging in to each instance iteratively, enter couple of parameters and then reports are generated. Not exactly a convenient practice.

  REM connect to each instance separately, type in the beginning snap_id and ending snap_id for each node etc..
   sqlplus mydba@proddb1
   @$ORACLE_HOME/rdbms/admin/awrrpt.sql
   exit;
   sqlplus mydba@proddb2
   @$ORACLE_HOME/rdbms/admin/awrrpt.sql
   exit;
   sqlplus mydba@proddb3
   @$ORACLE_HOME/rdbms/admin/awrrpt.sql
   exit;

There are few issues with this approach. It is a cumbersome practice if the instance count is higher. In addition to that, all of AWR reports are, in turn, accessing underlying AWR tables. Physically, rows from all instances are together in the same block and so, by executing these reports connecting to various instances, Global cache traffic is increased. If the database is suffering from Global cache (GC) performance issues then generating reports connecting to various instances is probably not a grand idea.

I have created few basic scripts to generate AWR reports from all instances in a single step. Script awrrpt_all_gen.sql queries AWR tables and gv$instance to access last snap_id and various instance_ids. Then this script generates AWR reports for all instances in that cluster in the current directory, for the last AWR snap. This script is quite handy while debugging RAC performance issues. Now, we have AWR reports from all instances in one directory and we can easily grep these files to understand metrics in various instances.

@awrrpt_all_gen.sql

awrrpt_all_gen.sql v1.00 by Riyaj Shamsudeen @orainternals.com

To generate AWR Report from all RAC instances concurrently.
Creates reports using last two snap_ids.

...Generating awrrpt_all.sql script.... Please wait....
...Completed script generation.

Executing awrrpt_all.sql to generate AWR reports.
...Generates AWR reports with file name format awrrpt___.txt for each instance.
...Please wait for few minutes...

...AWR report created for instance 1. Please wait..
...AWR report created for instance 2. Please wait..
...AWR report created for instance 3. Please wait..
...AWR report created for instance 4. Please wait..

AWR reports created.

If you would like ability to choose the snap_ids, then use the script awrrpt_all_range_gen.sql

Use granular approach to GC measurements

It is important to measure Global cache performance metrics correctly, especially those damned averages. Incorrect measurements of averages can lead to faulty analysis. For example, following query is retrieving average global cache CR receive time from gv$sysstat view. Problem with this SQL statement is that this statement is retrieving average global cache receive time from the restart of the instances. Global cache performance is quite sensitive to workload and querying over a wide time range, as in this query, can lead to faulty analysis.


REM This is really a bad idea!!

select b1.inst_id, b2.value "RECEIVED",
b1.value "RECEIVE TIME",
((b1.value / b2.value) * 10) "AVG RECEIVE TIME (ms)"
from gv$sysstat b1, gv$sysstat b2
where b1.name = 'gc current block receive time' and
b2.name = 'gc current blocks received' and b1.inst_id = b2.inst_id;

As an analogy, above query is equivalent to a patient-Doctor conversation:

Patient: Doctor, I don’t feel good. I feel that I am having fever.
Doctor : Nope. Your average body temperature from the beginning of this year is around 98. So, you are doing excellent.

It is essential to review the global cache performance metrics at a granular level. Use AWR reports or statspack reports for a short period of time. Better yet, if you want to quickly review GC performance metrics, you can use my script gc_traffic_print.sql. This script captures various statistics relevant to Global cache performance in to a PL/SQL table, sleeps for N seconds (default N=60) and re-captures the statistics after the sleep. Then the script prints all the relevant details in easily readable tabular format. Very useful in quickly reviewing the Global cache performance metrics.

SQL> @gc_traffic_print.sql

gc_traffic_print.sql v1.20 by Riyaj Shamsudeen @orainternals.com

...Calculating GC Rx and Tx timing and blocks..
...Default collection period is 60 seconds.... Please wait for at least 60 seconds...
Enter value for sleep: 30
---------|--------------|---------|----------------|----------|---------------|---------------|-------------|
Inst     | CR blocks Rx | CR time |  CUR blocks Rx | CUR time |  CR blocks Tx | CUR blocks Tx |Tot blocks   |
---------|--------------|---------|----------------|----------|---------------|---------------|-------------|
1        |         29288|    12.17|          14914|      4.58|          27059|           17439|        88700|
2        |         21201|     5.75|          12835|      5.19|          26831|           16741|        77608|
3        |         22942|     4.64|          11751|      3.82|          23918|            9205|        67816|
4        |         32075|     4.78|          19260|      4.55|          31046|           19331|       101712|
---------|--------------|---------|----------------|----------|---------------|---------------|-------------|

PL/SQL procedure successfully completed.

Column key for the above script is:
CR block RX : GC CR blocks received
CR time : Average GC CR receive time
CUR blocks RX: GC CUR blocks received
CUR time: Average GC CuR receive time
CR blocks TX: GC CR blocks transmitted
CUR blocks TX: GC CUR blocks transmitted
tot blocks: Sum of transmitted + received for both CR and CUR traffic

Breakup: global cache performance

Global cache processing performance can be broken down to its sub components ( excluding wire latency, but that can be calculated). Breakup of this metrics must be done at a granular level also. We can use yet another script gc_traffic_processing.sql to review the performance metrics for the past N seconds. For example, in the output printed below, Global cache CR processing time is broken in to various sub-components of global cache processing. Remember that these performance metrics are in the SEND side of Global cache performance metrics. Meaning, inefficient global cache processing in one instance affects other instances' global cache receive time, but not the self. For example, in the table above, GC CR receive performance was at 12.1ms for instance 1. That is due to another instance (probably node 4 in this case) suffering from high GC processing time (more specifically high 'CR block flush time' in node 4). It is far more accurate to review global cache performance issues in a granular basis. Caution though, try not to specify at least a minute so that these numbers are consistent.

@gc_traffic_procssing.sql
gc_traffic_processing.sql v1.00 by Riyaj Shamsudeen @orainternals.com

...Prints various timing related information for the past N seconds
...Default collection period is 60 seconds.... Please wait for at least 60 seconds...
Enter value for sleep:

---------|-----------|---------|-----------|----------|------------|------------|------------|----------|
Inst     | CR blk Tx | CR bld  | CR fls tm | CR snd tm| CUR blk TX | CUR pin tm | CUR fls tm |CUR snd tm|
---------|-----------|---------|-----------|----------|------------|------------|------------|----------|
1        |      62732|       .2|       1.25|       .21|       48920|        .96|         .22|        .22|
2        |      46037|      .13|        .76|        .2|       42019|        .68|         .17|         .2|
3        |      72237|      .63|        .79|       .19|       22697|        .44|         .29|        .23|
4        |     113816|      .13|       2.04|       .15|       51020|        .87|         .12|        .13|
--------------------------------------------------------------------------------------------------------

PL/SQL procedure successfully completed.

Here is the column definitions for the script output:
CR blk TX: GC CR blocks served
CR bld : Average GC CR build time/CR block served
CR fls tm: Average GC CR flush time/CR block served
CR snd tm: Average GC CR send time/CR block served
CUR blk TX: GC CUR blocks served
CUR pin tm: Average GC Current pin time /CUR block served
CUR fls tm : Average GC Current flush time/CUR block served
CUR snd tm: Average GC current send time/CUR block served

Review histograms for GC events

Averages can be misleading. Revealing performance issues underneath the average is a very important task to do. For example, we see that 'gc cr block receive time' average is high. But, let's review the histogram for the gc events related to Global cache receive time. My script gc_event_histogram.sql is quite handy to review event histograms. This script provides a difference in performance statistics between two samples and prints them in a readable format. In this case, node 1 is has 8578 waits for the event 'gc cr block 2-way' under 2ms and the sessions in instance 1 had 9851 waits between 2-4ms. Only few waits over 16ms.

@gc_event_histogram.sql

gc_event_histogram.sql v1.0 by Riyaj Shamsudeen @orainternals.com

Default collection period is sleep seconds. Please wait..
Enter value for event: gc cr block 2-way
Enter value for sleep: 60
---------|-----------------------|----------------|----------|
Inst id  | Event                 |wait time milli |wait cnt  |
---------|-----------------------|----------------|----------|
1        |gc cr block 2-way      |               1|        98|
1        |gc cr block 2-way      |               2|      8578|
1        |gc cr block 2-way      |               4|      9851|
1        |gc cr block 2-way      |               8|      2291|
1        |gc cr block 2-way      |              16|       117|
1        |gc cr block 2-way      |              32|         6|
1        |gc cr block 2-way      |              64|         1|
1        |gc cr block 2-way      |             128|         2|
1        |gc cr block 2-way      |             256|         0|
1        |gc cr block 2-way      |             512|         0|
1        |gc cr block 2-way      |            1024|         0|
1        |gc cr block 2-way      |            2048|         0|
1        |gc cr block 2-way      |            4096|         0|
1        |gc cr block 2-way      |            8192|         0|
1        |gc cr block 2-way      |           16384|         0|
2        |gc cr block 2-way      |               1|       155|
2        |gc cr block 2-way      |               2|     10792|
2        |gc cr block 2-way      |               4|     14201|
2        |gc cr block 2-way      |               8|      3887|
2        |gc cr block 2-way      |              16|       204|
...

But, we also need to see 3-way waits. Let's review 3-way waits with the above script. We will simply supply 'gc cr block 3-way' as the wait event. In this case, we see that there are 12,062 waits over between 4-8ms. Of course, this must be improved.


@gc_event_histogram.sql

gc_event_histogram.sql v1.0 by Riyaj Shamsudeen @orainternals.com

Default collection period is sleep seconds. Please wait..
Enter value for event: gc cr block 3-way
Enter value for sleep: 60
---------|-----------------------|----------------|----------|
Inst id  | Event                 |wait time milli |wait cnt  |
---------|-----------------------|----------------|----------|
1        |gc cr block 3-way      |               1|         3|
1        |gc cr block 3-way      |               2|       320|
1        |gc cr block 3-way      |               4|     17017|
1        |gc cr block 3-way      |               8|     12062|
1        |gc cr block 3-way      |              16|      1725|
1        |gc cr block 3-way      |              32|       113|
1        |gc cr block 3-way      |              64|         0|
1        |gc cr block 3-way      |             128|         0|
1        |gc cr block 3-way      |             256|         0|
1        |gc cr block 3-way      |             512|         0|
1        |gc cr block 3-way      |            1024|         0|
1        |gc cr block 3-way      |            2048|         0|
1        |gc cr block 3-way      |            4096|         0|
1        |gc cr block 3-way      |            8192|         0|
...

In essence, these four scripts will be useful in debugging Global cache performance issues. Please let me know if you encounter errors in these scripts. We will use these scripts in future discussions to show how we resolved few RAC performance issues.

RAC, parallel query and udpsnoop

I presented about various performance myths in my ‘battle of the nodes’ presentation. One of the myth was that how spawning parallel query slaves across multiple RAC instances can cause major bottleneck in the interconnect. In fact, that myth was direct result of a lessons learnt presentation from a client engagement. Client was suffering from performance issues with enormous global cache waits running in to 30+ms average response time for global cache CR traffic and crippling application performance. Essentially, their data warehouse queries were performing hundreds of parallel queries concurrently with slaves spawning across three node RAC instances.

Of course, I had to hide the client details and simplified using a test case to explain the myth. Looks like either a)my test case is bad or b) some sort of bug I encountered in 9.2.0.5 version c) I made a mistake in my analysis somewhere. Most likely it is the last one :-( . Greg Rahn questioned that example and this topic deserves more research to understand this little bit further. At this point, I don’t have 9.2.0.5 and database is in 10.2.0.4 and so we will test this in 10.2.0.4.

udpsnoop

UDP is one of the protocol used for cache fusion traffic in RAC and it is the Oracle recommended protocol. In this article, UDP traffic size must be measured. Measuring Global cache traffic using AWR reports was not precise. So, I decided to use a dtrace tool kit tool:udpsnoop.d to measure the traffic between RAC nodes. There are two RAC nodes in this setup. You can read more about udpsnoop.d. That tool udpsnoop.d can be downloaded from dtrace toolkit . Output of this script is of the form: