Re: RAC or Large SMP...?

From: <>
Date: Wed, 8 Oct 2008 16:24:54 -0700 (PDT)
Message-ID: <>

> If the point isn't x86 but rather running Linux then look at
> IBM's zSeries servers. You can have up to 54 IFLs with each
> IFL equivalent in horsepower to between 10 and 30 x86 cores.
> zSeries run zLinux with the best virtualization software, zVM,
> you can find anywhere. To find out more about Oracle on zLinux
> contact

I dont know too much about zSeries/Linux but I suspect the cost is very high when compared to xSeries.

I've been doing some research on the most powerful xSeries box available and it looks like we can get up to 64 cores by combining 4 xSeries 3850's (each with 4 quad cores) together.

We can run a single system image across the 4 chassis and this gives us the option of starting with one chassis and then adding more as the load increases (similar to adding a new RAC node to a cluster). The key benefits of this solution over RAC is that there is no need for the extra software layers (OCFS2, ASM, CRS, VIP, etc) or the extra RAC licensing cost. In addition we wont have the potentially large amount of data passing back and forth over the interconnect which could significantly limit scalability.

By combining multiple x3850 chassis together we will effectively have a NUMA based architecture, which means that there will be some memory access which has higher latency. However I see this as much less of a penalty that the overhead of RAC inter-instance traffic (both network delay and IPC overhead).

This doc explains the architecture pretty well I think:

I haven't made my mind up either way yet but I'm definately favouring the multiple x3850 idea over the multi-node RAC cluster.

Cheers Received on Wed Oct 08 2008 - 18:24:54 CDT

Original text of this message