RE: RAMSAN Experience

From: Mark W. Farnham <>
Date: Wed, 9 Sep 2009 16:56:59 -0400
Message-ID: <>

Read Tim's paper about scaling to infinity to see if you can apply his methodology to your constrained situation. (Partition exchange is a beautiful thing and Tim explains it better than most and how that in turn enables you to keep up with having the structures you need to give fast response. Very cool paper and if you can apply the concept to your situation you will be a hero, spend less money on hardware, and your CFO will direct your boss to give you a raise.

Greg's comments on price performance for persistent memory devices versus spinning rust are exactly on target. In your situation where you are trying to bust out TEMP space as an independent resource (if the Tim Gorman strategy doesn't work out for you {and since you're already partitioned on date it likely will work fine}), see if you can get your EMC savvy folks to actually do that for you (and make sure they give you enough cache along the way.) If they can and will, then you might be able to configure multiple users utilizing several independently operating TEMP[1,2,..n] tablespaces without collision such that the i/o actually streams. If they won't or can't, then even though the price/performance is a loser at scale, you *might* be able to get enough memory "disk" to fulfill your TEMP requirements at a price that makes sense in comparison to the entire existing diskfarm.

If I divide by 1000 correctly in my head, your tables are about 26G and 3G, so you probably do want to price out getting to 64bit hardware and loading up a big SGA.

And then there is always exadata, but your stuff seems a little small for that.

Good luck,


-----Original Message-----
From: [] On Behalf Of Rob Dempsey
Sent: Wednesday, September 09, 2009 11:29 AM To:
Subject: RE: RAMSAN Experience

Hi Tim / Greg

Thanks for the quick responses

I guess I should try and explain the problem a little bit more. This is a simplified version (it will make a much longer email to detail all our constraints and resources). We have a read only reporting application that allows users to query raw row level data. There are a number of combinations a query can have be it date period, type of products etc etc which makes it near on impossible for us to summaries the data - trust me I would if I could. For our large system the two main tables are in size

A	25690.25M
B      2955.25M

We use Oracle compression, pctfree 0, parallel query and partitioning on date. As our users seem to be addicted to response time being as low as possible and not having enough expertise in storage one solution was to set a db_cache_size that could accommodate all the table data and throw it into memory. Simply a large in memory database. This solution has worked very well for our smaller database, however as the data got larger hash joins, group bys are spilling to disk. PGA is set to a large value and my next point of call is to test different value for the underscore parameters that control it.

We use EMC storage however the latest idea is to use RAMSAM for the temporary files. I always thought it might be a good idea for the redo logs but I am not sure about the TEMP files.

Like I said we have a number of constraints, but any help would be welcome.


-----Original Message-----
From: Greg Rahn [] Sent: 09 September 2009 13:45
To: Rob Dempsey
Subject: Re: RAMSAN Experience

Out of curiosity, how large is your DW and what is workload today? Do you use parallel query?

I ask because SSD is not really that beneficial for a DW, especially when it comes to price/performance over spinning rust (HDDs). SSD is (mostly) about IOPS, not MB/s throughput which is where most DW are sized inadequately.

On Wed, Sep 9, 2009 at 4:54 AM, Rob Dempsey<> wrote:
> I have been asked to consider a RANSAN for our Oracle DW database.
> I was wondering if anyone had any experience of the technology? Pros and
> Cons.

Greg Rahn

Received on Wed Sep 09 2009 - 15:56:59 CDT

Original text of this message