[Ocfs-users] Re: Mode context extremely poor performance.
David McWhinnie
davidmcwhinnie at yahoo.com
Tue Nov 23 09:13:24 CST 2004
We had very similar problems, but was more pronounced
on writes instead of reads. Have you tried without
securepath? After working with HP and RedHat we found
there was problems in both the Kernel and SecurePath
when running OCFS.
We are running versions.
ocfs-2.4.18-e-smp-1.0.13-1
kernel-smp-2.4.18-e.47.6.5
gcc-2.96-128.7.2
fileutils-4.1-10.20
hp_qla2x00src-7.01.01-12 (qlogic driver)
Secure-Path-3.0CFull64-4.0RHEL2.1special
--- Aubertin Michael <maubertin at ares.fr> wrote:
> Hi all,
>
> Completary informations, we have already applied
> this patch:
> $ opatch lsinventory
> Installed Patch List:
> =====================
> 1) Patch 3849952 applied on Thu Nov 18 17:19:48 CET
> 2004
> [ Base Bug(s): 3849952 ]
> 2) Patch 3887769 applied on Tue Nov 09 18:01:09 CET
> 2004
> [ Base Bug(s): 3588448 ]
> 3) Patch 3802975 applied on Thu Nov 04 03:37:25 CET
> 2004
> [ Base Bug(s): 3553791 ]
>
> The issue is already the same.
>
> Thank you for your help.
>
> Athos10.
>
>
> Le lundi 22 novembre 2004 à 16:46 +0100, Aubertin
> Michael a écrit :
> > Hi all,
> >
> > I curently have a big problem. One request (listed
> above) using context
> > take up to 1000 more time than the on RAW or ext2
> database. I have ran
> > this request on a single IA32 machine with Redhat
> and dbf on ext2. The
> > average reponse time is less than a sec. The same
> request on RAC 4 nodes
> > cluster on RAW take the same average time. On ext2
> idem. But on OCFS it
> > took up to 15 sec randomly between
> 1sec>responce_time>15sec.
> >
> > We tried to ran the request on a single itanium
> (without RAC) on OCFS. -
> > > sameproblem.
> > We tries to ran the request on single nodes
> without RAC with only DBF
> > file on OCFS. -> Same problem.
> > We tried to run several version of OCFS -> same
> problem.
> >
> > We bench a few storage throuput with no revelant
> bottle neck. And in RAW
> > device the reponse time is perfect.
> >
> > I read OCFS source a recompile it in debug mode. I
> can see that the same
> > number of reading block demand (according to the
> sql log) follow an
> > increasing amount of ocfs_get_block2() call. In
> consequence the request
> > reponse time is dramaticaly increase. I tried to
> put more verbose debug
> > trace in ocfs_create_or_open_file() in order to
> grab to oin.hndl
> > releasing token, but i don't konw enough OCFS
> desing yet to be
> > performant, so i write to you to have help.
> >
> > First question: Could you help me please .....
> ;-).
> > Second: I see in the source you use VFS dentry
> struture. In the RH
> > 2.4.18 I2 kernel this structure was little
> modified to be more "2.6"
> > looked like. Is it cannot be the problem ?
> (Pointer to nowhere ? Gcc
> > BUG ?...)
> >
> > Thank you by advance for any help.
> > Athos10
> > aka
> > Michael Aubertin.
> >
> > Trace log available at:
> > -----------------------
> > http://athos10.dyndns.org
> >
> >
> > Software architecture:
> > ---------------------
> > Oracle 9.2.0.5
> > OCFS tested: 1.0.11 / 1.0.12 / 1.0.13 (all have
> the same problem).
> > Redhat AS 2.1 kernel e-41smp ITANIUM
> > gcc 2.96-128.7.2
> > fileutils 4.1-10.6
> > qlogic version 7.00.03 with HP register patch
> >
>
(ftp://ftp.hp.com/pub/softlib/software4/COL9063/co-24300-1/contrfailure.patch)
> > secure-path-3.0.cfull64-4.0.
> >
> > Hardware architecture:
> > ----------------------
> > 2 HP EVA3000 with Continious access real time
> replication.
> > 2 * 4 nodes HP RX4640 Itanium entry level
> servers.
> >
> >
> > SQL Request:
> > ------------
> >
> > SELECT COUNT(*) FROM SNAP.VWMT_PROFIL_MATCH_FH
> > WHERE CONTAINS (RECH_CRITERES, ' RZFR') > 0
> > /
> >
> >
> > The table:
> > ----------
> >
> >
> > SQL> desc snap.vwmt_profil_match_a_fh
> > Name Null? Type
> > ----------------------- -------- ----------------
> > ABO_ID NOT NULL NUMBER(11)
> > ABO_PHO NUMBER
> > ABO_PAYS VARCHAR2(4)
> > ABO_REGION NUMBER(11)
> > ABO_DEPARTEMENT NUMBER(11)
> > ABO_VILLE NUMBER(11)
> > ABO_CP VARCHAR2(50)
> > PROF_AGE NUMBER(4)
> > PROF_STATUT NUMBER(11)
> > PROF_TAILLE NUMBER(11)
> > PROF_POIDS NUMBER(11)
> > PROF_SILH NUMBER(4)
> > PROF_CHEVEUX NUMBER(4)
> > PROF_YEUX NUMBER(4)
> > PROF_ETHN NUMBER(4)
> > PROF_STYLE NUMBER(4)
> > PROF_REVENUS NUMBER(4)
> > PROF_ENFANTS NUMBER(4)
> > PROF_ENF_SOUHAIT NUMBER(4)
> > PROF_CAT_PROF NUMBER(4)
> > PROF_ETUDES NUMBER(4)
> > PROF_FUMEUR NUMBER(4)
> > PROF_NB_PHOTOS NUMBER(4)
> > PROF_DATE_CREATION DATE
> > PROF_DATE_MAJ DATE
> > PROF_NATIONALITE NUMBER(11)
> > PROF_LANGUE NUMBER(20)
> > PROF_ANNONCE NUMBER(11)
> > PROF_ANNONCE_VOC NUMBER(4)
> > AUTH_ACCES DATE
> > ANN_DATE_MAJ DATE
> > ABO_PSEUDO VARCHAR2(50)
> > PROF_LOOK NUMBER(4)
> > PROF_VIDEO NUMBER(4)
> > PERSO_ROMANTIQUE NUMBER(4)
> > PERSO_ATTIRANT NUMBER(20)
> > PERSO_MARIAGE NUMBER(4)
> > PERSO_CHEVEUXSTYLE NUMBER(4)
> > PERSO_RELIGION NUMBER(4)
> > INSC_DATE_INSC DATE
> > RECH_AGE_MINI NUMBER
> > RECH_AGE_MAXI NUMBER
> > RECH_TAILLE_MINI NUMBER
> > RECH_TAILLE_MAXI NUMBER
> > RECH_POIDS_MINI NUMBER
> > RECH_POIDS_MAXI NUMBER
> > RECH_CRITERES VARCHAR2(4000)
> > ZERANK NUMBER
> >
> >
> > The index:
> > ----------
> >
> >
> >
> > create index
> SNAP.CTX_PROFIL_MATCH_${ROLE}_${SEX}_01
> > on SNAP.vwmt_PROFIL_match_${ROLE}_${SEX}
> (RECH_CRITERES)
> > indextype is ctxsys.context PARAMETERS ('MEMORY
> 50M STORAGE
> > text_storage')
> > /
> >
> >
> > with:
> > -----
> >
> >
> > BEGIN
> ctx_ddl.create_preference('SYS.TEXT_STORAGE',
> > 'BASIC_STORAGE');
> ctx_ddl.set_attribute('SYS.TEXT_STORAGE',
> > 'I_INDEX_CLAUSE',
> > 'pctfree 0 tablespace TBS_SNAP_INDEX');
> > ctx_ddl.set_attribute('SYS.TEXT_STORAGE',
> 'I_ROWID_INDEX_CLAUSE',
>
=== message truncated ===
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
More information about the Ocfs-users
mailing list