End users will eventually adopt the iSCSI storage networking protocol, but many hurdles remain.
By Thomas Hammond-Doel
The storage industry's primary mandate is to ensure data is stored safely and retrieved accurately. Surrounding this mandate are all the disciplines, protocols, and implementations that pave the way. Fulfilling the mandate sets requirements on designs and implementations that cannot be compromised. Fibre Channel has met the challenges, and new protocols must also provide assurances of data integrity. One of the newest storage networking protocols is iSCSI, but there are many hurdles to overcome.
Numerous articles about iSCSI have been written over the last year. The terms "hype" and "reality" are bantered about to the point where end users may have difficulty discerning the truth.
One of the key questions is when iSCSI will overcome the obstacles that face any new protocol.
This article is not intended to promote or debunk myths, but rather to look at the iSCSI protocol realistically. With this examination coming from the Fibre Channel Industry Association (FCIA), comparisons to Fibre Channel will be presented.
IP was originally designed to support the US government in maintaining communications between R&D facilities, military bases, and other locations. Data frames were expected to be lost, and TCP was designed to reset to a known point to recover lost frames.
This focus led to an extremely flexible and persistent protocol. However, the original design concept for IP is in stark contrast to the design concept for Fibre Channel, where reliability was maximized and the foundation was based on the safe, guaranteed movement of data.
Most iSCSI implementations will require specialized host adapter cards for adequate performance. These I/O cards are variously referred to as iSCSI network interface cards (NICs) or host bus adapters (HBAs), and they all include TCP/IP offload engines (TOEs) to improve performance.
Several vendors are shipping, or are in the process of developing, pre-standard versions of these iSCSI cards. Each vendor implements TOEs differently, depending on where the protocol stack runs and whether micro-controller assists must be provided. Even with 30 years of maturity behind IP, iSCSI adapter vendors will still need to prove interoperability to the IT community.
iSCSI TOE cards are just beginning to cross the minefield of hardware and software bugs, which may affect data integrity in the short term. TCP ensures data will eventually get from point A to point B, but bugs currently add an element of uncertainty. Helping offset this potential problem is the maturity of TCP/IP infrastructure devices such as routers and switches.
Beyond interoperability issues, most iSCSI adapters will require significant buffering on the card, TOE, or motherboard (or a combination), which adds costs.
Using IPsec for security in an iSCSI implementation can also impose cost burdens. Currently, the overhead needed to meet IPsec requirements makes software implementations impractical.
IPsec ASICs are available, but they're expensive. If the iSCSI traffic is corporate data, the market for non-IPsec iSCSI adapters becomes questionable. The only ways around the security issue are to implement a separate infrastructure just for iSCSI traffic or to implement a security appliance.
Security appliances may help when extending beyond the firewall, but this adds to the total cost of ownership. Given that Fibre Channel has inherent security, currently runs twice as fast as iSCSI (over Gigabit Ethernet), and is more efficient at transporting storage data, the improved price-performance ratio of Fibre Channel may trigger a re-evaluation for certain applications.
Even with low-end applications where firewall security is sufficient, requirements for additional RAM will add cost to iSCSI adapters. With these added costs and limited target environments, iSCSI adoption rates may not cause quantity-of-scale price drops as quickly as previously expected.
When looking at the price-performance ratio in relation to hardware, early adopters of iSCSI storage area networks (SANs) have generally paid about the same as they would have for a Fibre Channel SAN, particularly in comparison to 2Gbps Fibre Channel SANs.
iSCSI adapter vendors may introduce cards at a price equivalent to—or slightly lower than—Fibre Channel HBAs, sacrificing margin for market share. This tactic may spur end-user adoption rates but will also prevent prices from dropping quickly.
For most iSCSI applications, Gigabit Ethernet speeds are required to meet a minimal level of performance. Depending on the complexity of a GbE LAN, the frame-loss ratio may cause performance to drop excessively for any application. And, while iSCSI currently runs over 1Gbps Ethernet, Fibre Channel can run at 2Gbps.
Even with typical GbE latencies and "best-case" frame-delivery ratios, iSCSI still suffers from "TCP droop," a condition where performance degrades during lost-frame recovery. This is especially true when iSCSI traffic runs on a network with non-iSCSI traffic.
IP was designed for sharing facilities among multiple applications. Synchronous applications, including some business continuance applications, do not run well on iSCSI. High-performance TCP applications can require as much as 50% to 75% over-allocation of bandwidth when adequate quality of service (QoS) is not available. This could limit iSCSI to low-end, asynchronous business continuance applications.
There are some business continuance applications that operate at acceptable efficiencies with iSCSI. However, a separate network may still be required for security and throughput. Nevertheless, these markets may provide considerable growth opportunity for iSCSI.
There is a perception that Fibre Channel is difficult to implement and understand, and there are few IT personnel who are knowledgeable about Fibre Channel. As some iSCSI implementers are finding out, the complexity comes from learning a new discipline (e.g., managing storage), not from learning a new protocol. A common adage is that it is easier to train a storage professional on IP than to train an IP professional on storage.
The concepts and disciplines involved with storage are not simple, and the tools to help make storage simple are still maturing. When end users do begin deploying iSCSI, they will benefit from the tools developed for Fibre Channel. Existing tools for IP are often inadequate for iSCSI storage.
It is highly unlikely that disk drive manufacturers will create native iSCSI devices, in part because of the TOE, IPsec chip, RAM, and support chips required, as well as form-factor and power-consumption issues. This leaves the front-end for iSCSI applications. Depending on the application, 10GbE may work for larger enterprises, but the initial cost-per-port of 10GbE is prohibitive for small or mid-sized companies today.
Small and mid-sized companies that have not already adopted Fibre Channel may see the attractiveness of using IP for SANs. As long as tools are developed to abstract the complexity of storage, this market has a lot of potential for iSCSI. Fibre Channel, on the other hand, will also benefit from the same tools, providing companies with a choice for storage networks.
The iSCSI specification is not yet finalized, and today's iSCSI products may not be compliant when the standard is ratified. This is typical in the genesis of a new protocol. After the standard is finalized, interoperability testing will begin (although iSCSI plugfests have already been conducted).
The iSCSI industry is establishing compliance programs similar to the FCIA's SANmark program, which helped Fibre Channel vendors attain interoperability through the development of test suites. Many of the meetings held for handling the interoperability issues of iSCSI include Fibre Channel vendors providing "lessons learned" and other advice.
The hurdles in front of iSCSI will eventually be cleared, interoperability will be attained, prices will drop, and the end-user community will adopt iSCSI. Ultimately, iSCSI and Fibre Channel will co-exist.
Over the last few months, several companies have announced delayed deployment plans for iSCSI. In a weak economy, enterprises are less likely to take chances on unproven technologies. Applications where iSCSI can play will become clearer over the next year.
Thomas Hammond-Doel is a board member of the Fibre Channel Industry Association (FCIA) and director of technical marketing at Vixel Corp.