Version & Recent Enhancement Information
The current version of SarCheck for AIX
is: 7.00.12
Enhancements to version 7.00.12 include:
(This is a beta for subscription holders only)
- Fix a typo in the fs1, ps1, and ps2 scripts
- Change references in SarCheck's text file from vmsparse to scaixagent
- Change parsing of lsps -a to allow for the syntax changes in AIX 6
- Check the validity of the data before reporting on paging space utilization
Enhancements to version 7.00.11 include:
(This is a beta for subscription holders only)
- Read filesystem & LVM data from /opt/sarcheck/fs
- Report on dio/cio mounted filesystems
- Create a separate disk table for EMC powerdisk disks
- Create fs1 script and fsaixagent
- Add -fs and -fsd switches to specify location of fsagent data
- Add FSDIR and PSELFDIR keywords to specify default fs & ps data directories
Enhancements to version 7.00.09 include:
(This is a beta for subscription holders only)
- Get MTU size from agent, include in analysis
- Start making recommendations for various networking parameters
- Fix minperm recommendation wording if os is 5.2 and pre52tune enabled
- Tweak decimal precision in table for network stuff
- Add a warning to the recommendations for maxrandwrt and j2maxRandomWrite that the likelihood of a performance improvement is application dependent
Enhancements to version 7.00.06 include:
(This is a beta for subscription holders only)
- Add parmsfile entries to suppress various tunable parameter recommendations
- Calculate default large process and memory leak thresholds from agent data
- Don't complain about small paging spaces if there's no (or trivial) paging activity
- Collect mtu data from netstat -in, non loopback interfaces
- Add parmsfile entries to customize the agent's search for MTU size
Enhancements to version 7.00.05 include:
(This is a beta for subscription holders only)
- Expand the explanation preceding the memory stats graph
- Simplify the y-axis scaling in the paging space activity, paging space used, cycle/s, and blocked I/O graphs
- Tweak wording of disk analysis if multiple %busy peaks are seen
- Change memory leak detection threshold to be 1/2% memory size if mem size is between 1-32gb
- Fix a bug in the cyclegraph when multiday data is analyzed
- If present, use clockhandrevs data in the cycle graph & text
- If > 3 clockhandrevs per interval, calculate mb/sec scanned
- Change (TM) to (R) or ® at the start of the report
- Update dkbusy graph to shows peaks as well as averages
- Fix runaway process & memory leak detection for multiday process data
- Remove "RS/6000" hardcoded in front of the model name
- Get memsize from rmss field in vmsparse output
- Support the analysis of data from one system on another
- Do some basic analysis of networking (no) parameters
- Add network stats to -t table
- Add support for parmsfile keywords WIDE PRP PML PLP DBRP DBML DBLP
- Add -newkey switch
- Tweak 2006 recs to prevent recommending maxperm 90->80, etc.
- Fix start/end reporting in multiday process (ps) analysis
- Change wording for bad sar -d data
- Increase max length of a -diag word
- Mention that j2maxrandwrt changes don't always work
- Use better decimal precision for sema/s and msg/s averages
- Fix timestamp problems in graphs of ps or agent data
- Change color of maxperm line on vmtune graph
- Make change decimal precision of cswch/s peak zero to match sar
- Make change decimal precision of %busy peak zero in text to match sar
- Fix HTML code if -html and no valid sar report found
- Calculate coefficient of correlation for cpu and %wio
- Fix code to allow multiple graphs to use multiple graph data files
- Fix the display of NOARP, NOTCP, NOUDP parms
- Add a -cs switch & CS parm to suppress cust settings
Enhancements to version 6.03.08 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Change vmsparse to collect numclust, j2_nPagesPerWriteBehindCluster, j2_nRandomCluster
- Change analyze to process numclust, j2_nPagesPerWriteBehindCluster, j2_nRandomCluster data
- Change vmsparse to collect maxrandwrt and numclust from vmtune as well as ioo
- Add recommendations for numclust, maxrandwrt, j2_nPagesPerWriteBehindCluster,
j2_nRandomCluster, j2_maxRandomWrite
- Clean up the code to characterize I/O
- Don't consider device names starting with cd* for disk balancing
- Recommend changing maxfree and minfree together
- Improve wording for %wio analysis
- Add clockhandrevs to cycle graph if the data is present
Version 6.03.07 was a beta test version
Enhancements to version 6.03.06 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Correct the problem of providing an incorrect example (for example,
vmstat vs. vmo) when making maxperm/maxclient/maxpin/minperm
recommendations
- Fix a problem parsing the output of oslevel when "oslevel -s" is used.
- Fix the message that the values of maxperm and minperm were 0.
- Enhance the disk table to only use Queue and Service Time columns when valid data is present in sar -d
- Correctly distinguish between maxperm and maxperm% metrics on newer operating systems
- Enhance warnings of likely problems when a sar report from one system is analyzed on another
Enhancements to version 6.03.05 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Fix a capacity planning bug, always saying memory will be used up first
- Fix a bug causing nonsensical maxperm and maxclient recommendations
- Rework parsing of vmsparse data to allow for better data analysis
- Display maxclient, memused, etc on vmtune graph
- Fix gnuplot errors if total paging space size is zero
- Reduced further the y-axis scaling on the ioblk graph
- Change alt text on vmtune graph
- Change description of the on vmtune graph
- Add avg value of numclient to -t table
- Add avg pct of memory used for non-file pages to -t table
- Add avg pct of total memory used to -t table
- Give the memory bar in the cap plan graph a tapered end
- Fix a problem with the -Q and -html switches working together
- Move legend outside the vmtune graph
- Change data & gnuplot scripts for cp and disk bar graphs
- Use a table with vmtune graph to show VMM stats
- Fix a bug in the "ignore short intervals" logic for sar -d data
Enhancements to version 6.03.04 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- The vmsparse agent now reports on iostat reads vs. writes
- Add LPAR name to key algorithm
- Check to see if our jfs/jfs2, seq/random write io guesses agree with the sys admin's thoughts.
This is a sanity check in preparation for better recommendations.
- Reduced minimum y-axis scaling on ioblk graph
- Rearrange pieces of the Resource Analysis section
- Fix a rare bug in the maxperm/minperm recommendations
- Fix a bug in printing informarion about the "2006 Tech University recommendations" if we're
not actually recommending changes
- Add % to vmo parameter names maxperm, minperm, maxpin, maxclient
- Fix the formatting of total pg space used date
- Changed wording that reported the analysis of multiday ps
- Add text for group of new vmo recs to be implemented together
Enhancements to version 6.03.03 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Fix a case where we weren't using the latest algorithms in the recommendations
- Fix parsing of lrupollinterval value
- Improve the internal algorithms for deciding if there's JFS and/or JFS2 filesystem activity
Enhancements to version 6.03.02 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Streamline some code
- Fix "Analysis not requested" msg if one of the
-ps switches is picked and no ps data file os found
- Discard minfree of less than 8 pages, not 10. A
recommended minfree change of 8 pages is not unlikely on
systems without JFS2 filesystems.
- Fix a typo in the text of a recommendation to change the
iostat attribute of device sys0
- Fix the precision used to display the value of "peak
swap queue" in the summary table
- Do a better job of handling bad data which shows up in
the sar -d report when sar's crontab entries have been
set up incorrectly
Enhancements to version 6.03.01 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- Change min/maxperm target (old pre-lru_file_repage
algorithm) allowed from 10 to 20
- Only consider pgspace < memsize a problem if there's a
memory bottleneck
- Remove "calculated entitled CPUs" from cpu2
graph
- If sys0 iostat=false, recommend setting it to true.
- Rewrite minfree recommendation algorithm for 5.3
- Discard minfree chgs of less than 10 pages instead of
less than 2 pages
- For low minfree values, recommend up to 4x increase
instead of limiting it to 2x
Enhancements to version 6.03.00 include:
(Subscription holders: If you'd
like this version, ask your sales rep)
- The following major changes came from our time at
IBM's 2006 system p Technical University:
- Add the new recommendations or lru_file_repage,
lru_poll_interval, minperm, maxperm, maxclient,
strict_maxperm/maxclient
- Use fsioblked data to make better numfsbufs
recommendations and add this data to the
"ioblk" graph
- De-emphasize the importance of %wio
- Change PARMS warning/error thresholds for WIOAVG from
20/50 to 60/80%
- Report on IO pacing parameters
- Analyze lsps -a data, intelligently scale the y-axis of
the "pgused" graph, and make recommendations
for paging spaces
- Some other enhancements:
- Remove text in Recommendations section about cpu &
disk if there are no recommendations for them
- Look for MB, GB, and TB in sys config line of sar data
- Fix an obsolete "Wrong OS" msg
- Print a note that %busy can be wrong if RAIDs are present
- Calculate average blks/IO to better characterize disk I/O
- Support the ever-changing blks/s, Kbs/s. and Kb/s column
in sar -d
- Don't mention system tables in some rarely seen text in
the Capacity Planning section
Back to the SarCheck home page
Copyright © 1996-2008 Aptitune
Corporation, All rights reserved.
Information in this document is subject to change without notice.
Other products and companies referred to herein are trademarks or
registered trademarks of their respective companies or mark
holders.
This information is correct as of May 13, 2008 and is subject
to change without notice.