Let me look into my crystal ball and see if we can define where the industry is heading over the next 18 months or so. I can see three major issues affecting the storage industry. They are:
- The management wars
- Virtualization standards
- The protocol wars
It sure has been an interesting year. We have seen major advances in standards on the fabric front and major fallouts and consolidation of certain segments of the market. Let's look at these to see if it brings to light what will be happening next year.
On the server front:
Whatever happened to Itanium and the 64-bit revolution? Server applications drive the need for storage. We have seen 64-bit Unix come to maturity, but what is lacking is the Wintel 64-bit side. The Itanium chip was released, but performance was not up to par due to chipset issues. The server vendors are hesitant to bring out models based on this design, and even if they do, where is 64 bit NT? Until Microsoft ships 64 bit Windows to the masses, and ISVs port their applications to utilize the benefits that 64 bits brings, this segment will stagnate. To have a balanced server architecture, Itanium servers will need a faster I/O bus. I can see this affecting the introduction of InfiniBand technology. The PCI bus has been around awhile, and has continually been improved. The problem is that the PCI bus is becoming the bottleneck when it comes to I/O. InfiniBand will solve this, but I don't see this happening until 64-bit processors drive the need.
On the protocol front:
ISCSI is slowly but surely coming along. Standards should come into place in 2002, and that will help drive iSCSI. I don't see iSCSI as a replacement for SAN fabrics yet though. We need to have more TCP/IP offload engines built into NIC cards, and 10GB Ethernet to take off before iSCSI becomes practical at the server level. iSCSI will first be used as a data replication protocol between SAN islands. Cisco will most likely drive this on the network side of things.
The real battle in data replication technology is whether to use iFCP or FC-IP for IP based data replication. iFCP is being driven by Nishan, and works seamlessly in letting devices in separate SAN islands communicate with each other over the link.
The name server information from the remote location is made available to the local location, and end nodes can communicate with each other using individual sessions over the link. This leads to very interesting ways to do remote replication. iFCP lets you tie two SAN islands together, while isolating local SAN traffic to the individual islands. Multiple sessions can be established over a single link.
FC-IP is more of an encapsulation protocol that allows FC messages to be packaged inside of IP frames. The SAN islands then become one big SAN tied together by the link. The link session is established between the devices used to connect the islands together. The fabric grows as you tie more islands together. This is also great for remote copy solutions, but it's not as flexible as iFCP in my opinion. Both will be standards and the market will decide who wins.
This was first published in January 2002