Enterprise storage systems are in for some major changes, as eWeek Labs Senior Analyst Henry Baltazar learned when he talked to some of the top players in the storage game. Chuck Hollis, of EMC Corp.; Jai Menon and Brian Truskowski, both of IBM; Network Appliance Inc.s Keith Brown; Randy Chalfant, of Storage Technology Corp.; and James Staten, of Sun Microsystems Inc., are unanimous in their belief that storage management is the most important area to watch in the future. Although they disagree about the prospects for IP storage protocols such as iSCSI, all believe these protocols will affect the storage networking landscape.
What technology will be the most important for storage managers to follow over the next few years?
The nearly universal answer for this question was that storage management was the hot technology area that IT managers should keep an eye on. It should be noted, however, that iSCSI and storage networking technologies came in a close second in almost all the responses.
Hollis, vice president of markets and products at EMC, said open storage management for switches, storage units and servers, regardless of vendor, must be established to lower the total cost of ownership of storage and simplify management.
Brown, Network Appliances director of technology and strategy, said he believes the DAFS (Direct Access File System) protocol will play an important role in data center storage infrastructures beginning next year. DAFS, a collaborative effort among dozens of vendors (including Network Appliance), will enable databases, Web servers, e-mail back ends and a host of other server-resident applications to achieve performance levels that are simply unattainable in the pre-DAFS world. (More information is at www.dafscollaborative.org.)
Menon, IBM fellow at the companys Almaden Research Center, touted IBMs upcoming Storage Tank storage management system, which he said combines storage virtualization, enterprise performance, policy-based storage management and data sharing across heterogeneous storage systems at a greatly reduced TCO due to more simplified management.
Staten, director of strategy for Suns Network Storage Division, and Chalfant, director of StorageTeks corporate strategy, indicated their companies lean more toward the storage virtualization and advanced volume management realm of storage management.
Although high-capacity storage and faster storage networking technologies are necessary, we believe management tools are the cornerstone for every storage implementation. And quite frankly, there is a lot of room for improvement in this field.
Where does iSCSI fit in the storage market?
Considering the high cost of fibre Channel HBAs (host bus adapters) and other Fibre Channel equipment and taking into account that the performance and capabilities of iSCSI arent known, eWeek Labs believes that iSCSI will affect the low end of the market first (see our April 9 Tech Analysis at www.eweek.com/links).
EMCs Hollis and Suns Staten concurred: Both predicted that iSCSI—at least in its initial introduction phase—would appeal mostly to the low-end (workgroup) segment.
Others didnt share this sentiment, however. Network Appliances Brown said that iSCSI fits into the midrange and high-end spectrum of the storage market.
According to Truskowski, vice president of technology and strategy in IBMs Storage Systems Group, iSCSI will first be deployed in the low-to-midrange market but will quickly move to the enterprise segment as technologies such as TCP/IP offload become more widely used. In addition, IP storage will simplify the complexity of SANs (storage area networks) while allowing many customers to use a networking infrastructure with which they are comfortable or at least have already deployed for other uses.
Chalfant predicted that iSCSI wont be limited to any single market segment; rather, it will be used by all market levels for business cases where storage traffic must travel over vast distances (Fibre Channel is limited to 10 km).
Do you believe SANs and storage networking in general will become predominantly IP-based in the future?
Network appliance and IBM are optimistic about IPs place in storage networking. (Network Appliances advocacy of IP is no surprise, given that the company uses IP to deliver file and storage services to clients and servers.)
“Looking back across the last 20 years, it would have been possible to place many, many bets against Ethernet and the IP protocol in the networking casino,” Network Appliances Brown said. “There is absolutely no reason to believe that a bet against IP and Ethernet technology today is any less risky than it has historically been.”
IP could be an important storage protocol, but this wont happen in the near future, according to EMCs Hollis. “The cost benefits of IP usually revolve around sharing of traffic with others. Any business requirement that demands deterministic performance will be unlikely to adopt IP-based solutions for these needs. There will be exceptions but not enough to displace SAN technology in the near future,” Hollis said.
Even if Ethernet hits 10G-bps speeds first, Suns Staten said, Fibre Channel will quickly follow. “A lot will depend on how efficient the two protocols [Fibre Channel and iSCSI] are. iSCSI, because of the TCP/IP overhead, may be less efficient and thus slower. But performance is not the determiner here,” he said. “Cost, complexity and stability are bigger issues that iSCSI looks to address.”
In short, although networking vendors (including Cisco Systems Inc.) are trying to shift the market with the expected emergence of the 10G-bps Ethernet protocol, performance (in terms of throughput per second) alone wont determine the status of IP storage protocols such as iSCSI. Latency, price/performance, reliability and distance limitations will determine which protocols will be successful in a given business case.
Will improved HBAs enable IP products to approach Fibre Channels performance and latency?
Latency is an extremely valuable metric, especially for environments with high transaction rates. SAN vendors often tout Fibre Channels low latency compared with IP. As Network Appliances Brown said, “Fibre Channel vendors are guilty of massively overblowing the performance and latency issues they say are inherent in IP networking.
“When coupled with the performance enhancements that can be derived by embedding file system and storage management/virtualization intelligence into storage appliances, the performance of IP-based storage products is often better than that of relatively dumb, block-based Fibre Channel devices,” he added.
IBM Almaden Research Centers Menon said he believes that iSCSI will create innovations and new features. Technologies such as TCP-offloaded HBAs and network processors in IP switches will enable IP storage to compete more favorably with Fibre Channel in performance, he said. Moreover, advancements in two other areas promise to give IP storage significant new advantages over Fibre Channel. Furthermore, IP storage will be able to deliver quality of service and security that is not currently available on Fibre Channel.
StorageTeks Chalfant said that although iSCSI will continue to improve, it would be unwise to discount Fibre Channel and Infiniband in the future. StorageTek doesnt really care which protocol takes control of the market because it will adjust its products to meet clients needs, he said.
TCP/IP offloading will help iSCSI, Suns Staten said, but its still too early to determine how well iSCSI will perform. “IP storages biggest problems will be handling the TCP/IP stack overhead and ensuring the correct ordering of packets. This will require cache and CPU cycles,” Staten said. “Initially, we expect to see the cache and CPU overhead be handled by the host, but then these burdens will move to the HBAs, which will provide greater cache and CPUs or ASICs [application-specific integrated circuits] for handling this work. It is unclear whether iSCSI will be able to achieve the same performance as Fibre Channel.”
EMCs Hollis said that with continued improvements, IP storage can eventually move into areas traditionally serviced by SANs.
Is Fibre Channel interoperability acceptable in its current form? As we move to 2G-bps and 10G-bps Fibre Channel, will interoperability issues among Fibre Channel components get worse?
“This is not a huge problem, and we dont expect it to get worse,” Truskowski, of IBM, said. “Vendor interoperability is improving. Industry initiatives are helping in this area. The real challenge is achieving interoperability at the customer API level, so operational procedures can be designed once, independent of which storage vendor it came from.”
Suns Staten held a similar opinion on the topic of interoperability but pointed out that true performance gains (moving up the ladder from 1G-bps to 2G-bps to 10G-bps speeds) cant be attained until all components move to those speeds.
Chalfant, of StorageTek, was far more negative in his assessment of Fibre Channel interoperability. “Interoperability and standards are an issue. The question in our mind is more like, Do you think the interoperability issues are acceptable to users? The answer to that is a resounding No! he said. “All vendors spend large sums of money in interoperability labs as a result.
“The chances of going out and buying bits and pieces and believing your SAN will work is near zero. You have to test this stuff to ensure there are no issues. Going forward, organizations such as SNIA [Storage Networking Industry Association] are having a large influence on standards. Eventually, it will all settle down, but it will be a few years at a minimum,” Chalfant said.
Network Appliances Brown was also quick to attack his companys Fibre Channel SAN foes: “Fibre Channel interoperability is definitely not acceptable in its current form,” he said. “In fact, Fibre Channel interoperability between vendors is almost nonexistent in its current form. The major Fibre Channel vendors have shown us all a very poor record in addressing the existing problems that have prevailed in the standard 100MB-per-second Fibre Channel space. It therefore seems highly likely that things will just keep getting worse as yet more Fibre Channel standards emerge.”
EMCs Hollis said the evolution of the Fibre Channel protocol wont make interoperability troubles any worse, but he recommended that IT managers should either dedicate resources to test Fibre Channel SANs internally or buy their solutions from vendors that have already done the required interoperability testing.
eWeek Labs believes that although interoperability has grown by leaps and bounds the past few years, Fibre Channel is still more patch and pray than plug and play.
What is the biggest challenge for centralized storage management?
Each vendor held different views on what must be done to improve centralized storage management.
Staten and Hollis said an open management platform must be created so all components, regardless of vendor, can be managed from a single point.
“Lack of good storage management standards and willingness by all vendors to cooperate [is the biggest challenge],” Staten said. “Sun … has been working with the FCIA [Fibre Channel Industry Association], SNIA and other organizations to promote open standards, yet many vendors continue to push for proprietary implementations that give them greater differentiation. The Ethernet market faced the same hurdles until standards emerged.”
Brown agreed that management of appliances and storage systems throughout a corporations global fabric was an important challenge but also suggested that replication and distribution of storage content were difficult issues, which Network Appliance is trying to address via its Data Fabric Manager and Content Director/Reporter products.
“The biggest challenge of centralized storage management is the fact that the data is typically written from many servers through many file systems, and there is not a single entity in the infrastructure that is managing this process,” IBMs Menon said. “Data is not placed on storage in consistent ways: The formats are all unique to the server file system that wrote the data, and there is no real thought given to efficient ways of placing that data.”
As networks get larger, how will your company help secure the data running through your storage network?
Until quite recently, security was an afterthought for most SAN vendors. Although an insecure SAN was somewhat tolerable early on—when Fibre Channel SANs were confined to server rooms and small test implementations on physically isolated networks—security cannot be an afterthought anymore. The upcoming expansion of IP storage networking will increase the exposure of data drastically, and SAN tools such as logical unit number masking and zoning are not enough to protect todays storage infrastructure.
“Fibre Channel was designed with minimal security features,” Menon said. “We have been working to improve security in SANs from the beginning.”
IBM helped develop SCSI-ACLs, a new security standard for storage networks (both Fibre Channel and IP) that was recently adopted by the T.10 technical committee of the National Committee on Information Technology Standards. (As the name implies, SCSI-ACLs add access control lists security to storage entities.) IBM is also researching ways to use encryption to better secure remote storage RAID arrays, Menon added.
Network Appliance will leverage IP security mechanisms for its storage networks. “Securing data on IP-based storage networks is a trivial task,” Brown said. “The mechanisms, technologies and products that provide authentication, encryption and VPN [virtual private network] services on IP networks are mature and well-understood by the industry and marketplace alike.
Sun is working with vendors such as Nishan Systems Inc. to secure IP storage networks by using standard IP firewalls, Staten said.
EMC is using a three-pronged approach: centralizing storage management, to eliminate the risks of having multiple storage management utilities active on a network; automating storage management tasks, to make storage less vulnerable to human error or data theft/corruption; and using hardware redundancy.
What is the most common problem customers have when they approach you for a storage solution?
Total cost of ownership seems to be the biggest driving factor for the implementation of new storage systems. Although direct-attached storage is still the norm, most companies are looking to centralize storage, according to the vendors eWeek Labs surveyed. Most of the respondents said their customers are fed up with the management difficulties and limitations of direct-attached storage and want a powerful, centralized storage system to simplify management and improve availability.