Re: file placement and SAN

Key questions for setting up on SAN:

How many spindles (sounds familiar)?

What size?

How many LUNS are presented back to the host?

How are they laid out on the physical spindles?

What is the stripe depth, i.e. how much data will be written on a spindle before moving to the next disk in the stripe set? db_file_multi_block_read_count should be a fraction of strip depth.

What is the size of a single I/O with the disk? Optimally, your block size will match this. If your block size is larger, will this trigger automatic read ahead?

What RAID level?

What about plaid (software striping over hardware RAID)?

Can you adequately express to your system administrators that storage capacity is not the issue, but that I/O capacity is the issue? For most databases when you have sufficient spindles and controllers to handle the I/O, you have more space than you need for storage. A single spindle can effectively handle 6-8 read and 1 write request at a time as a general rule of thumb.

If more than one host will be accessing the disk array, what is being done to ensure that only the database server can access the space allocated to it. In other words, your data cannot be accidentally mounted on another server and have newfs run on it (yes, this has happened to me).

How many hosts will be accessing the array? What is the total I/Os per sec that they willl be requesting? How about MB/sec?

Optical fiber is fragile, what is being done to ensure that if someone drops a floor tile on a cable, you do not lose access to the database (Veritas DMP for example)?

It is important to maintain the seperation of redo and archive logs from the datafiles, you should also seperate rollback and temp (in other words all the write intensive stuff). Are sufficient spindles being provided for this?

Cache is not a panecea for database I/O, most requests end up being for random reads. Writes to datafiles rarely can be built up into a full stripe, cache will not undo the RAID 5 penalty (to write a change to a single block, you must read the old parity block and the old data block, then write the new parity and data blocks).

On Mon, 29 Apr 2002 12:07:52 GMT, spamdump_at_nospam.noway.nohow (Ed Stevens) wrote:

>On Thu, 25 Apr 2002 18:15:44 +0100, "Paul Brewer" <paul_at_paul.brewers.org.uk>
>wrote:
>
>>"Ed Stevens" <spamdump_at_nospam.noway.nohow> wrote in message
>>news:3cc72643.124146312_at_ausnews.austin.ibm.com...
>>> SELECT
>>> feedback,
>>> comments,
>>> experiences,
>>> FROM
>>> newsgroup_participants
>>> WHERE
>>> san_experience not null;
>>>
>>
>>Try again, without the comma after 'experiences', and you may get some data
>>returned ;-)
>>
>>HTH,
>>Paul
>>
>Well, I would like to report a bug, wherin the above query should return an
>ORA-00936: missing expression, but instead returns sporadic results. However,
>it is no longer possible to talk to a human being on Oracle's support line, and
>our purchasing department has let our support contract lapse so that I cannot
>open an i-tar on MetaLink. Just as well, they (Oracle) would probably tell me
>that the version of Oracle I'm running isn't supported on the brand of ethernet
>cable we use.
>
>Let's try this one. Perhaps it will also prevent including NAS_FEEDBACK,
>NAS_COMMENTS, and NAS_EXPERIENCE from the result set.
>
>SELECT
> san_feedback,
> san_comments,
> san_experience
>FROM
> newgroup_participants
>WHERE
> san_experience not null;
Received on Mon Apr 29 2002 - 15:39:34 CDT