Exacheck reports "FAIL Storage Server Check Exadata Smart Flash Cache should be actually used"

From: Yong Huang <"Yong>
Date: Tue, 27 Oct 2015 14:22:11 -0700
Message-ID: <1445980931.75817.YahooMailBasic_at_web184806.mail.gq1.yahoo.com>



On Exadata X4-2 running Oracle 11.2.0.4, smart flash cache seems to be unused. Some tables have cell_flash_cache set to KEEP and most are left as DEFAULT and I also tested by explicitly creating a few GB table with cell_flash_cache set to KEEP and did some DML and queries.

The database stats:

SQL> select inst_id, name, value from gv$sysstat where name like '%flash%' and value > 0;

   INST_ID NAME                                                                  VALUE

---------- ---------------------------------------------------------------- ----------
1 cell writes to flash cache 5514015 2 cell writes to flash cache 2961032

So there're writes to flash cache. Careful checking of gv$sesstat indicates they're all done by LGWR. (We have another Exadata machine and I ran the above query on it and got both that stat and "cell flash cache read hits" as expected.)

On the cell server, command "list metriccurrent FC_BY_USED,FC_BYKEEP_USED" shows 0. "list metriccurrent where metricObjectName='FLASHCACHE'" shows 0 for all lines except for

 FC_IO_BY_W_SKIP FLASHCACHE 16,485,257 MB 
 FC_IO_BY_W_SKIP_LG FLASHCACHE 15,010,155 MB 
 FC_IO_BY_W_SKIP_LG_SEC FLASHCACHE 0.019 MB/sec 
 FC_IO_BY_W_SKIP_SEC FLASHCACHE 0.108 MB/sec 
 FC_IO_RQ_W_SKIP FLASHCACHE 146,704,160 IO requests 
 FC_IO_RQ_W_SKIP_LG FLASHCACHE 35,947,762 IO requests 
 FC_IO_RQ_W_SKIP_LG_SEC FLASHCACHE 0.2 IO/sec 
 FC_IO_RQ_W_SKIP_SEC FLASHCACHE 7.8 IO/sec 

Command "list flashcachecontent attributes cachedkeepsize,cachedsize,dbid,dbuniquename,hitcount,misscount,objectnumber" returns nothing. "list metriccurrent fc_by_used" still shows 0.

Flash cache disks have been dropped and recreated. It made no difference.

What could be preventing flash cache from being used (except by LGWR)? Thanks.

Yong Huang
--

http://www.freelists.org/webpage/oracle-l Received on Tue Oct 27 2015 - 22:22:11 CET

Original text of this message