Disclaimer 1: I work with FlexPods (Cisco UCS+VMware+NetApp), vBlocks (Cisco UCS+VMware+EMC) and other Cisco UCS+NetApp designs.
Disclaimer 2: These comments are my own and not those of my employer, customers or business partners.
I went and put together a short comparison "if I were to buy one of the two" without even going to software features integration with VMware vCenter and/or vCloud.
So, this is what I found:
Now, from my perspective, if I'm designing a cloud service or a virtualization infrastructure right now, the first protocol I will look is FCoE (performance and unification to the max) together with NFS (cost effective and fast way of providing many advanced features in VMware) without leaving FC (there still some performance scenarios where I'll stick with it) and iSCSI (when using certain 10GE cards and proper encoding, it outperforms FC and is way more cost effective). Only not having FCoE, for me, it is a major drawback for 3PAR.
I am into the unified storage protocol infrastructure. That is the way to go. It has happened with telephony and collaboration and now it is happening to storage thanks to virtualization. It makes datacenter life so much simpler. (But this is not what this post is about).
Deduplication is another feature that is a win-win for datacenter and virtualization.
With the higher consolidation rations that are possible nowadays not having 10GE support is another major drawback. How will you maintain a decent iSCSI throughput over 1GE?!
You may found that back in 2007, 3PAR and NetApp had some collaboration agreement. From what I see, basically it was 3PAR disks with NetApp brain (NetApp V-Series). Kinda when Toyota puts their chassis to NASCAR cars but the engine and everything else is something else. Is it the cassis or the engine that count for the race?
3PAR Multi-Tenancy and Thin-Provisioning support seems to be very decent. NetApp does provide these two features as well. Implementation details might or might not impact performance or storage efficiency or scalability, I don't know. I still have to read more to be able to technically compare them, so, for now, I will say both have these two feature.
Why is this important? These two features are really key for any virtualization infrastructure. When doing virtualization you don't want to over-provision, you want your ROI fast and juicy. At this point, I won't even consider virtualization without thin provisioning. At least in NetApp, there is no performance degradation and it is a major advantage against other competitors.
Seems HP is is working in a path to develop a new 3PAR architecture with modern hardware support. That will be very good. Some analysts are talking about a something based in the EVA family with 3PAR technologies. Lets say a mix of the two. That will be interesting. Will that happen? We should know by the end of 2011. FCoE will be a must. They should also consider deduplication capabilities and thin replication.
I believe that relaying in FC disks and not supporting SAS disk is holding them back. FC disks are expensive and they production is in decline. When that happens market tell you that prices will continue rising. Haven't we've seen it with petroleum?
Even more, from the IDC and Gartner 2010 report as summarized by http://export.imix.co.za/node/93872:
Gartner's market research projects that 4 Gbps Fibre Channel will fall to 10.1% this year and to 5.6% in 2011 en route to extinction in 2014. At that time, unit shipments of SATA (50.1%) and SAS (49.9%) will run neck-and-neck, and the SAS throughput speed will increase to 12 Gbps, according to Gartner.
IDC predicts vendors will discontinue shipments of Fibre Channel HDDs by 2013, a year earlier than Gartner's forecast. John Rydning, research director for HDDs at IDC, attributed the shift from Fibre Channel to SAS to vendors' desire to get to a common architecture. He said economies of scale will help to push down the overall cost of SAS components.
So, when it comes to virtualization and cloud services only having support for old expensive hard disks, only support 2x storage protocols, requiring physical Windows servers and no being able to support the newer datacenter trends, it does not look like a fit for me. I must agree that it was good at some point (2004 - 2007) but not anymore. Unless drastic changes are done to the architecture, it will just be "another one".
Is NetApp for all virtualization and cloud scenarios? I don't think so. Does it help with most of them? Certainly.
NetApp should start looking into higher dedup optimization in LUNs. They should also look into supporting higher FCoE ports concentrations and put more effort into the management GUI.