need help for Informix 5.10 slowness in I/O

need help for Informix 5.10 slowness in I/O

Post by Chen, Ale » Wed, 03 May 2000 04:00:00



Our Informix 5.10 under Sequent ptx 4.4 has encountered slowness issue. The
major cause is due to the disk I/O contention. Below is the output of the
command, sar -d 10, as you can see that the contention occurred in disk area
(data: sd02010, sd02011, sd02012, sd02013, and mirror: sd02110, sd02111,
sd02112, sd02113). This was the result from consolidation of the large table
extends into one extend, which is a good practice recommended by Online 5.
However this practice causes the disk I/O contention! What is the best way
to manage the data distribution? We have stored the data in the separated
dbspace from the rootdbs. Is there anyway to adjust the data among the disks
since 5.10 doesn't support the table fragment?

Another option is to increase the BUFFER number (currently set as 28000).
What's the best number?  The application system here doesn't use the
transaction. So LRUS = 32, LRU_MAX_DIRTY = 8 , and LRU_MIN_DIRTY = 3. They
appear not good. What's the best figures to reduce the disk I/O? Your prompt
suggestion will be appreciated.

Alex.
703-762-5020
===

DYNIX/ptx numaq 4.0 V4.4.4 i386    05/01/00

10:26:35  device   %busy   avque   r+w/s  blks/s  avwait  avserv

10:26:37  sd01001       3     1.0       3      41     0.0    11.7
          sd01002       4     1.0       3      18     0.0    13.3
          sd01003       3     1.0       3      41     0.0    10.0
          sd02000       2     1.0       2      17     0.0     8.0
          sd02001       5     1.0       4      32     0.0    12.2
          sd02002       2     1.0       2      12     0.0    12.5
          sd02003       7     1.0       4      26     0.0    15.6
          sd02010      95     5.0      82     896    46.2    11.6
          sd02011      99    15.3      82     828   172.7    12.1
          sd02012      91     4.4      79     808    38.7    11.4
          sd02013      99     6.0      79     859    62.7    12.5
          sd02100       2     1.0       2      16     0.0    10.0
          sd02101       4     1.0       4      32     0.0    10.0
          sd02102       3     1.0       3      28     0.0    10.0
          sd02103       1     1.0       1       8     0.0    15.0
          sd02110      89     4.9      80     772    43.9    11.2
          sd02111      92     6.2      81     892    58.7    11.4
          sd02112      94     5.3      88     856    45.7    10.7
          sd02113     100    13.4      84     777   147.1    11.9

 
 
 

need help for Informix 5.10 slowness in I/O

Post by Art S. Kage » Wed, 03 May 2000 04:00:00


Alex, you need to post just a bit more information.  This does look like
disk contentionbut whether there is an Informix solution or an OS one is
not clear yet.  Please post:

tbstat -P
tbstat -d
tbstat -D
tbstat -F
tbstat -R
tbstat -u

and the length of time since the last time tbstat -z was run before these
reports were taken if after startup.  Also what kind of drives are these
(ie 1GB, 2GB, etc.; manufacturer & interface[SCSI-II, SCSI-III, etc.]).

Art S. Kagel


> Our Informix 5.10 under Sequent ptx 4.4 has encountered slowness issue. The
> major cause is due to the disk I/O contention. Below is the output of the
> command, sar -d 10, as you can see that the contention occurred in disk area
> (data: sd02010, sd02011, sd02012, sd02013, and mirror: sd02110, sd02111,
> sd02112, sd02113). This was the result from consolidation of the large table
> extends into one extend, which is a good practice recommended by Online 5.
> However this practice causes the disk I/O contention! What is the best way
> to manage the data distribution? We have stored the data in the separated
> dbspace from the rootdbs. Is there anyway to adjust the data among the disks
> since 5.10 doesn't support the table fragment?

> Another option is to increase the BUFFER number (currently set as 28000).
> What's the best number?  The application system here doesn't use the
> transaction. So LRUS = 32, LRU_MAX_DIRTY = 8 , and LRU_MIN_DIRTY = 3. They
> appear not good. What's the best figures to reduce the disk I/O? Your prompt
> suggestion will be appreciated.

> Alex.
> 703-762-5020
> ===

> DYNIX/ptx numaq 4.0 V4.4.4 i386    05/01/00

> 10:26:35  device   %busy   avque   r+w/s  blks/s  avwait  avserv

> 10:26:37  sd01001       3     1.0       3      41     0.0    11.7
>           sd01002       4     1.0       3      18     0.0    13.3
>           sd01003       3     1.0       3      41     0.0    10.0
>           sd02000       2     1.0       2      17     0.0     8.0
>           sd02001       5     1.0       4      32     0.0    12.2
>           sd02002       2     1.0       2      12     0.0    12.5
>           sd02003       7     1.0       4      26     0.0    15.6
>           sd02010      95     5.0      82     896    46.2    11.6
>           sd02011      99    15.3      82     828   172.7    12.1
>           sd02012      91     4.4      79     808    38.7    11.4
>           sd02013      99     6.0      79     859    62.7    12.5
>           sd02100       2     1.0       2      16     0.0    10.0
>           sd02101       4     1.0       4      32     0.0    10.0
>           sd02102       3     1.0       3      28     0.0    10.0
>           sd02103       1     1.0       1       8     0.0    15.0
>           sd02110      89     4.9      80     772    43.9    11.2
>           sd02111      92     6.2      81     892    58.7    11.4
>           sd02112      94     5.3      88     856    45.7    10.7
>           sd02113     100    13.4      84     777   147.1    11.9


 
 
 

1. OnLine 5.10 w/CSDK & ESQL 5.10

A while back I asked about running D4GL with OnLine 5.10 and CSDK.  
Art and Jonathan pointed out that I-STAR would be needed with that
mix of products when the ESQL/C version is above 5.10.

But Jonathan also wrote that:

Has anyone succeeded with this technique?  Any ideas how I'd go about it?

--

Washington Group International, Inc.         (215) 422-4144
Philadelphia, PA, USA
Any opinions I state are my own and not necessarily those of my employer

2. Accessing Indexes with DAO

3. Informix 5.10 and VPN Connection

4. drop column doesn't work in oracle 9i

5. Fail Consistency Check AIX & Informix Online 5.10

6. Php question Php Questions !!

7. Informix 5.10 configuration

8. Failing Stored Procedures

9. Problems with Arcunload & Informix 5.10

10. Informix Online 5.10 and Legato Networker

11. INFORMIX 5.04 & 5.10 on Solaris

12. Informix -254 errors with JAM DBi 5.10

13. Help dumping Btrieve 5.10 files to ascii