Thanks for the question, Srinivas.
Asked: January 26, 2017 - 12:30 pm UTC
Last updated: January 26, 2017 - 10:31 pm UTC
Version: Oracle 11g R2
Viewed 1000+ times
You Asked
Dear Oracle Team,
We are building a ODS (Operational Data Store) which will be used to generate Reports.
The ODS contains 6 schemas dedicated to each of the Source databases.
1. Size of data on Deployment date: 500 GB (for all 6 schemas).
2. We have cut-over time of 6 hours to load all 500 GB data.
3. DMLs : 20 per sec/Schema, which makes 100 MB per schema/month, populated by a replication tool.
4. Selects : 50 per sec/schema, i.e 300 concurrent sessions each will fetch 50KB of data.
5. Reports: 50 reports will be generated Hourly with 20MB data for each.
Considering above, Could you please help with me the sizing...
How much RAM, CPUs and Cores would be required, with some justification.
Thankyou,
Srinivas
and Connor said...
You need about |-------------------| this much :-)
Jokes aside, thats not a good way at all to decide on how much RAM, CPU etc you need. One report could be:
select sum(..) from one_table
whereas another report could be:
select [hundreds of calcuations]
from [25 table join]
My view is why take a "guess" when you can "know", ie, probably get a much better estimate by either
a) renting some cloud space and do some realistic benchmarking with your data, or
b) using a subset of data on spare capacity on a non-production server and doing some benchmarking.
Then you'll have some real data - which even if it turns out to be incorrect, you'll have that solid evidence of your testing/benchmarking to fall back on.
Otherwise, if you pick 8 cores and it turns out you needed 16 cores, then saying "Someone on AskTom, who has never worked with our systems, said 8 cores" probably wont look too good to your management :-)
Is this answer out of date? If it is, please let us know via a Comment