A little known benefit – IP
SANcastle is lumped together with the 487 other smaller convergence switch players that are fighting to tie geographically remote SAN islands together. What separates SANcastle from the pack is something that the Fibre Channel guys ought to be bragging about: A SANcastle box can effectively offload server TCP/IP (a hot topic today!) in a Fibre Channel environment.
If a user has a Fibre Channel SAN, all the servers in that SAN are also attached to Ethernet. Inter-node communication and server-to-client communication both cause the dreaded pig known as TCP to rear its ugly head. By sticking a relatively cheap SANcastle box in the middle, users can offload the TCP processing down to the Fibre Channel HBA and effectively eliminate 80%-plus of the cycles that server required for TCP.
This scenario directly translates to better application and server performance and can dramatically reduce Ethernet network traffic and congestion. No one seems to know about this, but anyone running a Fibre SAN today will benefit so dramatically, it will be impossible to not justify the minor expense.
The real beauty is that nothing else in the environment needs to change – it just makes the whole darned thing work better. Since no one sold TCP performance as a benefit when they justified their Fibre Channel purchase, this is all gravy.
Steve is right about this one. It's only a matter of time before the Fibre Channel vendors move in the direction of accomodating server-to-server traffic on the SAN. Right now customers are still growing into their SANs so it will take time and probably a stable budget period or two before they can be expected to move from the basic server-to-storage messaging to include storage-to-storage messaging and finally, to include server-to-server messaging as their SANs expand and as new applications are developed that take advantage of the deterministic (read: predictable) environment of the SAN. In so many words, any rapid customer uptake in this area will decisively end the 2nd generation SAN interconnect war and set the early stage for the 3rd generation SAN interconnect war that already has 10Gbps as the natural convergence point for FC, Infiniband and Ethernet.
The natural reference for this progression can be found in the first generation SAN environment (ESCON) since the only way to communicate with a mainframe is still through the dynamic point-to-point connections created by the ESCON (FICON) directors. All print servers, file servers, application servers and storage devices (disk and tape) continue to use the half duplex ESCON interconnect (full duplex FICON interconnect) to communicate with the mainframe or clustered mainframes. Effectively, the ESCON directors (and FICON directors) and adapters can already handle server-to-storage, storage-to-storage and server-to-server traffic.
The difference between the mainframe SAN environment and the open system SAN environment, of course, is that the mainframe architecture has effectively remained in the hands of one vendor (IBM) for more than 30 years so it has more mature parallelism and operating performance (sustained to peak loads) characteristics than multi-vendor open systems.
20+ year-old Unix, for example, has fragmented over the years into multi-vendor memory hogs that still do not have the same type of parallelism or peak performance levels as mainframes.
NT/W2K is only 10+ years old and it is going through what promises to be a gruelling 32-bit to 64-bit transition period along with Intel's own transition from 32-bit CISC to 64-bit EPIC. As the saying goes, it generally takes Wintel software application at least 3 full versions to fully exploit the increased power of the hardware.
What these all mean, of course, is that unlike the homogeneous mainframe environment, the risk-averse customers will most probably move more gradually from homogeneous server-to-server messaging before moving fully into heteregeneous server to server messaging on the SAN.As the joke goes, open systems mean maximum choice which also means maximum confusion.<g>
My sense is that this evolution will generally trail the way customers are migrating their applications from the mainframe environment to the Unix environment to the NT/W2K environment. Like the way that porcupines mate, they continue to do it very, very slowly because while applications ultimately drive infrastructure sales; changes in business processes do tend to lag changes in software which, in turn, tend to lag changes in hardware.
In the Wintel platform, for example, software generally tends to improve at a 10% rate a year while hardware generally tends to improve at a 30% to 35% rate a year. Storage networking technically narrows this inherent gap between Wintel hardware and software and gives it a faster track into the data centers that one would expect given the relative immaturity of this unified platform. It's entirely conceivable that, thanks to networked storage, I/O-challenged NT/W2K will probably do unto Unix what was done unto the Mainframe by Unix if you accept the premise that Linux is just another Unix variant that does very well at refining settled ground but does not do too well in taming unsettled ground -- always a touchy point to make.
Anyway, most companies are typically deploying SANs and NAS before or after their LAN upgrades (generally from 100Mbps to 1Gbps) with the goal of ultimately tying together those SANs and NAS in an enterprise-wide SAN such as that exemplified by McDATA's 7,000-port deployment SAN deployment which consists of multiple director-based backbones that are just now branching out into switch-based department and workgroup SANs. As many here know, CMNT and INRG provide the best way to follow the SWAN component of those enterprise-wide deployments.
Right now, some insights can be gleaned from the way customers are deploying the following:
1) Stand-alone NAS -- tactical way to deploy and consolidate general purpose file servers.
2) SAN with hybrid SAN/NAS front-end -- strategic way to start deploying and consolidating all types of storage. Note the server consolidation and application consolidation happening in conjunction with the dynamic pooling of storage.
3) SWANs plus SANs with hybrid SAN/NAS front-ends -- connects all the networked storage in the enterprise to provide a consistent view of the data.
A good example is what appears to be happening at EMC which has products that span all three phases of deployment. Its NAS business increased by more than 80% from 3Q01 to 4Q01 indicating to me that instead of instead of buying more Symmetrix, its large installed SAN base continues to fill out its networked strategy by deploying lower-cost Celerra Highroad, Chameleon and software. Most of the CMNT and INRG SWAN business is still coming its channel extension installed base -- typically the largest companies in the world -- with some business coming from the surviving xSPs. |
