<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:taxo="http://purl.org/rss/1.0/modules/taxonomy/" version="2.0">
  <channel>
    <title>topic Re: cpu 100% slow reads fast writes in Operating System - HP-UX</title>
    <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233940#M890667</link>
    <description>Do you have some stupid sqls running ? I found them many in my envs !! as k the application guys to re-write the codes !&lt;BR /&gt;&lt;BR /&gt;When you have a bad I/O I would determine which table space is being used the most from the dba's , find out the file , and then the LV and stripe it to a number of disks if possible !&lt;BR /&gt;&lt;BR /&gt;Here I think your node is running out of resources in terms of mem,copu and IO time to plan for an upgrade !!&lt;BR /&gt;&lt;BR /&gt;Kaps</description>
    <pubDate>Wed, 31 Mar 2004 03:15:50 GMT</pubDate>
    <dc:creator>KapilRaj</dc:creator>
    <dc:date>2004-03-31T03:15:50Z</dc:date>
    <item>
      <title>cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233934#M890661</link>
      <description>My server is showing cpu utilization at 100%, I know that is not a bad thing, but it seems like when jobs are running that require reading to be done they take a very long time, but jobs that are writing run very fast.  The disk utilization is averaging 30-40%.  Memory is around 80%, and swap is constant at 46%.  It is an oracle database that we are running no this server.  Does anyone have any suggestions about what else I may want to look at.</description>
      <pubDate>Tue, 30 Mar 2004 10:54:51 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233934#M890661</guid>
      <dc:creator>Derek Baxter_2</dc:creator>
      <dc:date>2004-03-30T10:54:51Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233935#M890662</link>
      <description>What are your settings for dbc_max_pct and dbc_min_pct? What do you see in sar -b 5 5?&lt;BR /&gt;&lt;BR /&gt;Is %rcache high? What is the RAM on machine?&lt;BR /&gt;&lt;BR /&gt;How the file system is created? Post mount -p|grep "file_system_on_which_reads_are_slow"&lt;BR /&gt;&lt;BR /&gt;Anil</description>
      <pubDate>Tue, 30 Mar 2004 11:01:58 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233935#M890662</guid>
      <dc:creator>RAC_1</dc:creator>
      <dc:date>2004-03-30T11:01:58Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233936#M890663</link>
      <description>It is recommended for an Oracle Database to mount filesystems where datafiles are with "mincache=direct,convosync=direct" options. It bypasses the OS disk cache (Oracle does its own).&lt;BR /&gt;&lt;BR /&gt;You should also have a look at dbc_min_pct and dbc_max_pct kernel parameters, which determines the size of the OS IO cache.</description>
      <pubDate>Tue, 30 Mar 2004 11:04:28 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233936#M890663</guid>
      <dc:creator>Fred Ruffet</dc:creator>
      <dc:date>2004-03-30T11:04:28Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233937#M890664</link>
      <description>If you want to know if there a problem with your database, searching it with system tools is just not a straightforward approach.&lt;BR /&gt;&lt;BR /&gt;If the RBDMS is Oracle :&lt;BR /&gt;To get a general idea&lt;BR /&gt;set linesize 200&lt;BR /&gt;set pagesize 200&lt;BR /&gt;set time off&lt;BR /&gt;set timing off&lt;BR /&gt;&lt;BR /&gt;COL SID                 FORMAT 99999&lt;BR /&gt;COL SEQ#                FORMAT 9999999&lt;BR /&gt;COL EVENT               FORMAT A28      TRUNC&lt;BR /&gt;COL P1TEXT              FORMAT A20      TRUNC&lt;BR /&gt;COL P1                  FORMAT 999999999999&lt;BR /&gt;COL P1RAW                               NOPRINT&lt;BR /&gt;COL P2TEXT              FORMAT A20      TRUNC&lt;BR /&gt;COL P2                  FORMAT 999999999999&lt;BR /&gt;COL P2RAW                               NOPRINT&lt;BR /&gt;COL P3TEXT              FORMAT A20      TRUNC&lt;BR /&gt;COL P3                  FORMAT 999999999999&lt;BR /&gt;COL P3RAW                               NOPRINT&lt;BR /&gt;COL WAIT_TIME           FORMAT 9999999  NOPRINT&lt;BR /&gt;COL SECONDS_IN_WAIT     FORMAT 9999999&lt;BR /&gt;COL STATE               FORMAT A10      NOPRINT&lt;BR /&gt;&lt;BR /&gt;select * from v$session_wait&lt;BR /&gt;/&lt;BR /&gt;&lt;BR /&gt;select EVENT,count(*)&lt;BR /&gt;from v$session_wait&lt;BR /&gt;group by EVENT&lt;BR /&gt;order by count(*)  desc&lt;BR /&gt;/&lt;BR /&gt;&lt;BR /&gt;To analyze the results, look at wait events explaination in Oracle Performance Guide.&lt;BR /&gt;&lt;BR /&gt;For a fast I/O surevillance :&lt;BR /&gt;set pagesize 200&lt;BR /&gt;set timing off&lt;BR /&gt;set time off&lt;BR /&gt;&lt;BR /&gt;SELECT &lt;BR /&gt;        ses.sid&lt;BR /&gt;        , DECODE(ses.action,NULL,'online','batch') "User"&lt;BR /&gt;        , MAX(DECODE(sta.statistic#,9,sta.value,0)) /greatest(3600*24*(sysdate-ses.logon_time),1) "Log IO/s"&lt;BR /&gt;        , MAX(DECODE(sta.statistic#,40,sta.value,0)) /greatest(3600*24*(sysdate-ses.logon_time),1) "Phy IO/s"&lt;BR /&gt;        , round(60*24*(sysdate-ses.logon_time),0) "Minutes"&lt;BR /&gt;FROM &lt;BR /&gt;        V$SESSION ses&lt;BR /&gt;        ,V$SESSTAT sta&lt;BR /&gt;WHERE &lt;BR /&gt;        ses.status = 'ACTIVE'&lt;BR /&gt;AND &lt;BR /&gt;        sta.sid = ses.sid&lt;BR /&gt;AND &lt;BR /&gt;        sta.statistic# IN (9,40)&lt;BR /&gt;GROUP BY &lt;BR /&gt;        ses.sid, ses.action, ses.logon_time&lt;BR /&gt;ORDER BY&lt;BR /&gt;        SUM( DECODE(sta.statistic#,40,100*sta.value,sta.value) )&lt;BR /&gt;/&lt;BR /&gt;&lt;BR /&gt;&lt;BR /&gt;You could also install statpack, it will report a lot of usefull information.&lt;BR /&gt;&lt;BR /&gt;</description>
      <pubDate>Tue, 30 Mar 2004 11:13:08 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233937#M890664</guid>
      <dc:creator>Nicolas Dumeige</dc:creator>
      <dc:date>2004-03-30T11:13:08Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233938#M890665</link>
      <description>I like the reply from Nicolas best: If you like to understand what Oracle is doing, then just ask Oracle what it is doing! Often with a cursory glance at first, and a detailed statspack later.&lt;BR /&gt;&lt;BR /&gt;What does the IO subsystem look like? Many (most?) systems these days have physical disks behind an intellegent and caching controller. The writes to such controller as often cached into its (battery backup protected) caches and are acknowledged instantaneously. The reads however need to go on to the disk (unless re-reading or read-ahead is workginf): do the seek, wait for the rotation to come around to the right place, start the transfer to the controller, then transfer to the host. So write are then fast (typically 1 milli second) while reads are slows (typically 5 ms, but 15 is not uncommon).&lt;BR /&gt;To add insult to injury, if the reads and writes for to the same (set of) disks then those write will go and 'disturb' the reads by causing more and largers seeks. And you can queue up many more writes (in the same time) as reads! &lt;BR /&gt;Here... try a mental experiemnt: Assume a trivial synchroneous programs. One issues a write, wait for teh result and issue the next. The Other issues a read, waits for results and issues the next. Well, in the time the reader gets teh first IO back (5 ms?), the writter will have done 5 writes, and those 5 writes will be queued up to the real disk and 'in the way' of the 2nd  read request which will now take 5ms for a current write to finish + 5 ms for a queusd up write to finish and then finally 5 ms for the read to finish. By now the write has done 10 more IOs, all stashed in the cache, waiting to hurt the reader some more. It's a good thing that eventually even writer have to read what to do next! :-).&lt;BR /&gt;&lt;BR /&gt;waddyathink, makes sense?&lt;BR /&gt;&lt;BR /&gt;Hein&lt;BR /&gt;&lt;BR /&gt;</description>
      <pubDate>Tue, 30 Mar 2004 13:09:00 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233938#M890665</guid>
      <dc:creator>Hein van den Heuvel</dc:creator>
      <dc:date>2004-03-30T13:09:00Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233939#M890666</link>
      <description>hello, it could also be that the database is very fragmentated we had that problem too, after full export dropping database and doing an import it was much better.&lt;BR /&gt;the database then is "new" have a look that the oracle file are big enough at beginning and the extends are big enough . look if you have some hotspots in your disks, if one disk is writing and reading much more than others.&lt;BR /&gt;</description>
      <pubDate>Wed, 31 Mar 2004 01:28:29 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233939#M890666</guid>
      <dc:creator>Ingo Duschl</dc:creator>
      <dc:date>2004-03-31T01:28:29Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233940#M890667</link>
      <description>Do you have some stupid sqls running ? I found them many in my envs !! as k the application guys to re-write the codes !&lt;BR /&gt;&lt;BR /&gt;When you have a bad I/O I would determine which table space is being used the most from the dba's , find out the file , and then the LV and stripe it to a number of disks if possible !&lt;BR /&gt;&lt;BR /&gt;Here I think your node is running out of resources in terms of mem,copu and IO time to plan for an upgrade !!&lt;BR /&gt;&lt;BR /&gt;Kaps</description>
      <pubDate>Wed, 31 Mar 2004 03:15:50 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233940#M890667</guid>
      <dc:creator>KapilRaj</dc:creator>
      <dc:date>2004-03-31T03:15:50Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233941#M890668</link>
      <description>It does look like it might be something in the database.  I appears that is only taking a long time reading from one table, which is the one we use the most.  We are trying to work with that table to see if we can speed it up.  Thanks for the help.</description>
      <pubDate>Wed, 31 Mar 2004 09:05:55 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233941#M890668</guid>
      <dc:creator>Derek Baxter_2</dc:creator>
      <dc:date>2004-03-31T09:05:55Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233942#M890669</link>
      <description>&lt;BR /&gt;Install the Enterprise Manager and run the SQL analyzer.  You can see what your top SQL statements are based on disk I/O, buffer gets, etc, and look at the explain plan for those statements.  Chances are, you need an index or your database is not using the indexes that are there.  If you're running 8.1.7 or above, are using the COST or CHOOSE optimozer mode, and have gathered statistics on your tables and indexes, you may need to adjust some things like OPTIMIZER_INDEX_COST_ADJ.  By default, the optimizer in COST mode (there is no RULE optimizer mode in 9i) says that indexes are 100% as "expensive" as a full table scan.  Set the OPTIMIZER_INDEX_COST_ADJ in the init.ora file to "5".  This will tell the optimizer that an index is 95% lower cost than a full table scan.  Improves performance greatly.  This works, of course, provided that you have indexes on the table that the optimizer can use for the query you're running.&lt;BR /&gt;&lt;BR /&gt;-greg</description>
      <pubDate>Wed, 31 Mar 2004 11:23:37 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233942#M890669</guid>
      <dc:creator>Greg OBarr</dc:creator>
      <dc:date>2004-03-31T11:23:37Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233943#M890670</link>
      <description>For the fragmentation hypothsesis&lt;BR /&gt;&lt;BR /&gt;If the table is updated / deleted with hight frequency, it can get fragmented. To check this out : &lt;BR /&gt;&lt;A href="http://www.remote-dba.cc/oracle_tips_chained_rows.htm" target="_blank"&gt;http://www.remote-dba.cc/oracle_tips_chained_rows.htm&lt;/A&gt;&lt;BR /&gt;&lt;BR /&gt;If so you can export, drop, then import the table to get the best physical allocation.&lt;BR /&gt;&lt;BR /&gt;If the table is very large, you can use partition and spread the IO on disks.  &lt;BR /&gt;&lt;BR /&gt;Indexes, materialized view ... Oracle give you a lot of option to get better performance.&lt;BR /&gt;&lt;BR /&gt;</description>
      <pubDate>Wed, 31 Mar 2004 12:13:03 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233943#M890670</guid>
      <dc:creator>Nicolas Dumeige</dc:creator>
      <dc:date>2004-03-31T12:13:03Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233944#M890671</link>
      <description>You could be seeing a large number of parsing of SQL statements as well.&lt;BR /&gt;&lt;BR /&gt;Check your settings for session_cached_cursors and cursor_sharing.  &lt;BR /&gt;&lt;BR /&gt;Setting session_cached_cursors from 100-300 should be good.  If you are using bind variables, cursor_sharing should be set to SIMILAR.  If you are using literal statements, you can set it to FORCE.  There is some performance hit for it, however this might help keep the parsing down on your SQL statements. &lt;BR /&gt;&lt;BR /&gt;Thanks,&lt;BR /&gt;&lt;BR /&gt;Brian</description>
      <pubDate>Wed, 31 Mar 2004 16:59:33 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233944#M890671</guid>
      <dc:creator>Brian Crabtree</dc:creator>
      <dc:date>2004-03-31T16:59:33Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233945#M890672</link>
      <description>Hi,&lt;BR /&gt;&lt;BR /&gt;For me I will find the top process that hightly utilize CPU , if this is Oracle Server process , we can check for SID and SQL that currently running and take huge CPU.&lt;BR /&gt;from OEM. Or v$session and v$sql like:&lt;BR /&gt;&lt;BR /&gt;SELECT DISTINCT /*+ORDERED*/ &lt;BR /&gt;sql.sql_text, &lt;BR /&gt;sql.address, &lt;BR /&gt;sql.hash_value, &lt;BR /&gt;0 piece, &lt;BR /&gt;sid, &lt;BR /&gt;serial#, &lt;BR /&gt;getusername(sid)username &lt;BR /&gt;FROM v$session s, &lt;BR /&gt;v$sql sql &lt;BR /&gt;WHERE sql.address=s.sql_address &lt;BR /&gt;AND sql.hash_value=s.sql_hash_value &lt;BR /&gt;&lt;BR /&gt;---------&lt;BR /&gt;Then try to analyze table that related in the query using&lt;BR /&gt;dbms_stats package.&lt;BR /&gt;&lt;BR /&gt;HTH</description>
      <pubDate>Thu, 01 Apr 2004 07:24:54 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233945#M890672</guid>
      <dc:creator>Printaporn_1</dc:creator>
      <dc:date>2004-04-01T07:24:54Z</dc:date>
    </item>
    <item>
      <title>Re: cpu 100% slow reads fast writes</title>
      <link>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233946#M890673</link>
      <description>Thanks to everyone for the help.  I rebuilt the table and it has fixed the problems.</description>
      <pubDate>Thu, 01 Apr 2004 09:03:51 GMT</pubDate>
      <guid>https://community.hpe.com/t5/operating-system-hp-ux/cpu-100-slow-reads-fast-writes/m-p/3233946#M890673</guid>
      <dc:creator>Derek Baxter_2</dc:creator>
      <dc:date>2004-04-01T09:03:51Z</dc:date>
    </item>
  </channel>
</rss>

