VLM and putting an entire database in memory

With an Alpha GS140 and gigabytes of RAM, you can apparently use something called VLM to put an entire database into memory.

However, when I'm looking at a database said to be configured this way (I didn't configure the hardware or the database - Oracle 8.0.5 on OpenVMS), I see the database buffer cache hit ratio hovering around 50%; a far cry from the 95-99% figure said to be desirable.

The shared pool is around 30% utilizied and is about 150MB. No obvious problem there.

But the database buffer cache is only set to be 5MB! (five). So now I'm confused. Might the hit ratio be grossly skewed if the "entire database" is "in memory"? Any reason why someone wouldn't increase the DB_BLOCK_BUFFERS if memory were plentiful? Also noted these values:

db_block_size = 2048
db_file_multiblock_read_count = 16

Isn't 2048 bytes ridiculously low? Shouldn't it at least be doubled for a large production system (not that I can rebuild the entire database at the moment)? What about multiblock_read_counts? Any reason it shouldn't be doubled to 32?

Thanks.

• Dana

Sent via Deja.com
http://www.deja.com/ Received on Thu Dec 14 2000 - 14:28:34 CST