raw or Hp-Ux file's for I/O

[someone_4]

Here is an email I got today. What information are the requesting? I dont even know where to start on this one.

Richard

Is web1 using raw or Hp-Ux file's for I/O? If it is using file system for I/O we can modify the kernel to free up memory by modifying the buffer cache..

[Craig Rants]

I think a good analogy would be looking at how dba's want to use disk space as opposed to regular use. dba's prefer raw space, not going through /opt/oracle/ora01... but using /dev/dsk/rc0t0d0, stuff like that.

I hope I have understood their question and answered appropriately.

C

[someone_4]

I thought it had something to do with the db. but not sure not too much detail there.

[S.K. Chan]

Is web1 using raw or Hp-Ux file's for I/O? If it is using file system for I/O we can modify the kernel to free up memory by modifying the buffer cache..

Usually raw vs fs IO refers to database installation, whether it is using a raw disk or filesystem as storage. For example Informix prefers raw disk access for speed. Look at all your lvols (except swap) and see if all of them are filesystem lvols. In an Informix environment for example, a raw lvol used by the database does not hv a fs mounted and it's device file (eg: /dev/vg02/rlvol2) belongs to Informix UID.
The kernel in question is bufpages which is actually defined as (NBUF*2)

[Roger Baptiste]

I guess web1 is either the name of a system or database on the system. The next guess is the mail asks whether the database is using RAW volumes or filesystems.
This can be checked by doing a bdf on the system and see whether there are any database mountpoints. If it is not, also do vgdisplay -v and see whether there are raw volumes. The DBA can aid in this simply by looking at his end!(poking in the database).

The kernel parm they are talking about is the buffer cache (bufpages, dbc_max_pct and dbc_min_pct ).

HTH
raj

[James R. Ferguson]

Hi Richard:

There are certainly a handful of things that need to be known to begin to provide meaningful responses to this one.

Are you memory constrained? How much memory do you have and what pressures is it experiencing?

Is this a database server, like Oracle, where the database software manages its own buffers?

Do you have I/O bottlenecks?

Regards!

...JRF...

[Sridhar Bhaskarla · Accepted solution]

Unless we know what's web1, it will be difficult for us to come to a conclusion.

If web1 is a server, you can find it is using raw or file system i/o by doing a

#sar -b 2 500

Observe the output for sometime. You are interested in looking at pread/s and pwrit/s. This is the raw I/O. If it is constantly 0, then you are not using raw I/O.

If web1 is a database or having a database, you can ask your DBA to see if they are using raw devices for the database.

What the mail might be saying is to check the dbc_max_pct and dbc_min_pct values. They are percentages. Check the values and calculate the actuals by percenting with the physical mem. If your dbc_max_pct (default 50%) is set to 25% and if you mem is 4 GB, you are using 1 GB for buffer cache which may be waste of memory and you may want to reduce it to 10%. It's advisable to keep the buffer cache below 500MB. So you can calculate the percentage based on your Physical memory.

-Sri

[Henrique Silva_3]

This is an easy thing to miss if you are a sys admin that do not build the devices for application managers, such as dbas.

What we have done, since I have both dba and sys admin experience, is to ask the sys admin to give me access to a box during database install, especially of sybases prior to version 12, and establish a naking convention for the raw devices that would be obvius to anyone.

so, instead of naming a raw device /dev/vg01/rlvol1 or wahtever the default is, you could name it /dev/vg01/sybase_data1 or /dev/vg01/oracle_data01, etc.

That way, it is obvious that this is a device that means something to someone, and you do not have the problem of a junior sys admin to blow your database away by mistake here.

I hope this helps,

Henrique Silva