In this TR, Oracle workloads are compared on FCP, iSCSI and NFS.
SPOILER ALERT: FCP wins. (Kind of)
For total IOPS at 1ms latency, this is what the graph looks like:
From the TR:
In Figure 1, the difference between FC and DNFS is about 10%. However, this number should not be interpreted to mean that placing a given workload on DNFS generally results in a 10% decrease in performance. Not all database performance problems result from storage I/O latency. Furthermore, the difference in performance shown in Figure 1 is at a very high I/O level. Lower IOPS levels show a smaller difference between protocols.
While FCP is likely going to be faster than NFS, the key point of the TR is that Oracle + Flash is faster than traditional storage:
From the TR:
Because of the inherent latencies associated with spinning media, SAS-based HDD storage systems cannot achieve latencies better than approximately 8ms. Although differences exist between the protocols, any protocol with AFF offers an approximately tenfold improvement in performance.
What does this mean for NFS?
Absolutely nothing. Sure, FCP was faster in these tests. But it’s not a block vs. file thing. It’s a “you get what you pay for” thing.
FCP comes at a cost – you pay for HBAs, switches, licenses, etc. With NFS, you pay for a license and leverage your existing network. Is that added cost really worth a potential 10% boost in IOPS? Maybe. But it’s definitely worth thinking long and hard about it.
Be sure to take a long look at this TR and make your own judgment, but one thing is clear – if you’re running Oracle, you want an all-flash solution. All protocols win on flash.