Webster defines architecture as a unifying or coherent form of structure. SAN means BLOCK data transfers on a specialized network while NAS means FILE data transfers on the general purpose ethernet network. The integration of SAN and NAS means that BLOCK and FILE data transfers can occur on the specialized network and the general purpose network according to application requirements.
Sun invented NAS in the 80s using a general purpose Unix server as a file server to address the 80/20 networking phenomenon: 80% of the traffic on a network segment consists of local traffic while only 20% of the traffic is external. Thus, overall network and application server performance could be improved by offloading the file serving, print serving and later, mail server functions to specialized local servers.
Auspex and Network Applicance took that one step further in the early 90s by developing native embedded operating systems dedicated solely for file serving, removing the overhead of a general purpose Unix server and later, the general purpose NT servers. That thin server approach remains the conventional design for NAS today.
EMC diverged from Auspex and Network Appliance by developing Celerra which is an adaptation of video server technology. Here's a 1996 column by Bob Metcalf that puts this technology in perspective:
infoworld.com
Netapp and Auspex did not consider Celerra as competition even though Celerra went from $7M in revenue in 1Q98 to $210M in revenue in 4Q00. This is not surprising. NTAP and Auspex were evangelizing the NAS-only approach to
storage while EMC, based on its experience with ESCON-based mainframe SANs and its 1995 acquisition of McData, already saw the potential of Fibre Channel to integrate SANs and NAS.
The common complaint was that EMC was also counting very expensive Symmetrix disks in Celerra revenue. This cosmetic complaint was often made by NAS-only advocates who were fighting a losing war because they could not see the complementarity of SAN and NAS. In the process, they put themselves in a curious position whereby they were asking for the disk portion of a storage solution to be excluded from the revenue count!!!
Fortunately, that tiresome SAN vs NAS debate has been mooted by the entry of IBM, HWP and HDS/NNI with Celerra type NAS heads which validate EMC's approach which, in retrospect, was clearly ahead of its time. The research houses also see the complementarity of SAN and NAS. IDC is already using NIS (networked information storage) in parallel with their conventional SAN and NAS forecasts. Gartner/Dataquest is already using FAS (fabric attached storage) also in parallel with their conventional SAN and NAS forecasts.
EMC started referring to their SAN and NAS revenues as networked information storage in the 4Q00 earnings release. Prior to that they classified networked storage revenue in a variety of ways reflecting the ambiguities of an emerging market. Early on they highlighted the amount of revenue from systems with fibre channel connectivity then they started using SAN/NAS highlights then ESN highlights before finally settling on networked information storage or NIS. Not coincidentally, they started doing this after the release of the Chameleon NAS and Celerra HighRoad.
So the direct answer to your question about double-counting is that yes, there is necessary double-counting because nobody buys a Symmetrix for file-serving only. There is no practical way to divide the total cost of a Symm based on the disks dedicated to Celerra, the disks dedicated to OLTP and the disks dedicated to OLAP especially with more dynamic configuration features built into the Symm all the time. But those rigid SAN and NAS classifications were always artificial classifications to begin with so that distinction is not all that important. Going forward, NIS is the key category for networked storage. SAN-only and NAS-only strategies are losing strategies. Period. |
