Re: Direct NFS and ZS3

From: Mladen Gogala <>
Date: Thu, 07 May 2015 01:07:05 -0400
Message-ID: <>

On 05/06/2015 03:02 PM, Kevin Jernigan wrote:
> Mldaen,
> The primary design goal for Direct NFS (dNFS) was / is to provide
> SAN-equivalent (or better) performance, in terms of both latency and
> throughput, while using NFS / Ethernet infrastructure. A secondary
> goal is to simplify the configuration and tuning process for Oracle
> Database with NFS storage. dNFS accomplishes these goals by
> implementing the NFS client inside Oracle Database, rather then using
> the OS-supplied "kernel" NFS client. This allows dNFS to skip some
> parts of the networking stack, and to skip some of functionality that
> is required for a general-purpose NFS client, such as write ordering.
> In addition, dNFS creates a separate connection to the NFS server for
> each Oracle process, unlike kNFS, which essentially multiplexes all
> the processes' I/O's through one or a very small number of connections
> to the NFS server. There are other optimizations in dNFS which provide
> major performance improvements over kNFS, and which allow dNFS to
> auto-configure itself based on interrogating the NFS server.

Well, I'll have to check it out then. I was lazy and was simply using /etc/exports and put the nfs4 mount with the corresponding options into /etc/fstab and that was about it. So far, I used kNFS because I know how to configure it and have been using it for a long, long time. Usually, I didn't have much trouble with it, if the connection was fast enough. 10GB LAN is the norm for NFS based DB.

> In general, if the NFS server can handle the workload, then dNFS can
> provide SAN or iSCSI-equivalent performance, with very little
> configuration work required of the DBA or system administrator.

In my experience, NFS provides better performance than iSCSI without a specialized HBA. As I have said, I don't have numbers, just impressions.

Mladen Gogala
Oracle DBA

Received on Thu May 07 2015 - 07:07:05 CEST

Original text of this message