Issue TOC Print full article Print this page Download as PDF E-Mail this URL     In this article Our Scenario Company RFIs Cover Story: Nice, Neat Storage: The Reality

Compaq Computer Corp.

Compaq sent us a whopper of a proposal, which can be summed up as "Trash your storage network, and start over with ours." It'll cost you a hair over $5 million. This solution was the biggest, most detailed and most expensive in the bunch.

The only "full-service" vendor in our group, Compaq responded with a plethora of hardware, software and services. The solution replaces EDC's existing infrastructure with Compaq's own equipment and includes platinum support with four-hour response time, 24 hours a day, 365 days a year. Compaq boasts, rightfully, of its long history of protecting its customers' hardware investments, and its equipment is expandable to handle increased data capacities. Problem is, Compaq's virtualization scheme requires the Compaq hardware, yet a forklift upgrade defeats the purpose of virtualization. Furthermore, much of the proposal deals with benefits you could get from implementing Compaq's SAN solution rather than its virtualization. If we were EDC, we would forklift this proposal.

Compaq's Proposed SAN Click here to enlarge

Compaq takes an out-of-band approach to virtualization. The company proposed its VersaStor architecture, which features virtualization data written on a chip in each Fibre Channel controller on the network. Compaq's SANworks Virtual Replicator application runs on a Microsoft Windows 2000 platform but is not strictly required for the SAN to be virtualized. All you need for continuous uptime is data-path integrity, which Compaq's hardware makes possible through the use of multiple redundant paths. Compaq provides mirroring functions between the data centers via its StorageWorks Bi-Directional Data Replication Manager. The company didn't address our question regarding the capacity of EDC's existing circuits to sustain the data rate for its mirroring solution.

Compaq offered the widest range of services for installation and configuration of the hardware and software. The vendor's service organization and professional services are both highly developed. The other vendors' plans for installation would likely prove adequate, but none is as integral to the big picture as Compaq's proposal.

Reporting, trending and planning are important. For our RFI, Compaq referred us to its 14-page white paper "Business Value Methodology in Support of Networked Storage Architectures". Professional and complete, this document outlines a philosophy for creating policies and procedures for a SAN or virtualized SAN. No other vendor provided anything like it.

Compaq sent a long dissertation of the ROI benefits of its package, including a projected ROI time of 16 months. The vendor also submitted a 16-page analysis of EDC's situation from ITCentrix, which says that ease of management stems in part from a reduction in storage staffing. We hope that no one on EDC's storage team would read the ROI portion of the proposal. Compaq certainly provided the figures and tools to help convince management that its $5.06 million solution is best. However, the discussion did not account for the losses from the disposal of EDC's current storage equipment.

Compaq's proposal had a number of strengths. Clearly, the writer put serious time and thought into the project and provided the most comprehensive set of ROI calculations we received. The vendor's descriptions and rationales were complete. Furthermore, factors such as stability, longevity and reputation give Compaq an edge if we had to start from scratch. Even with the company's merger with Hewlett-Packard Co., we are not concerned about the fate of Compaq's storage operations, which are a jewel in the company's crown. Most likely, HP's storage division will play second fiddle to Compaq's. Still, for EDC, it makes no sense to start from the ground up when reasonable upgrade paths are available for far less than $5 million.

Compaq Computer Corp., (800) OK-COMPAQ; www.compaq.com

DataCore Software Corp.

If we awarded a bid on completeness and sensibility alone, DataCore would win. DataCore submitted a well-crafted response with a detailed, phased implementation plan and a reasonable cost of $868,356, justified by a pricing summary. We liked the level of data the vendor provided and that DataCore told us when it was making an assumption. Also, we appreciated the time DataCore spent on the phone with us, making sure it could fulfill our requests. That kind of commitment bodes well for the company's prospective customers.

But completeness alone can't win this contest, because any technology could wind up as a dead end. We worried, for example, about DataCore's reliance on Fibre Channel in a world that is moving full force into an IP environment. DataCore's suggestion that the path to new transport technologies is the use of transport transformation routers is really a duck under the question of future upgradability. We would like DataCore to be more flexible in its choice of technologies instead of relying on other vendors to provide it with router/converter devices.

DataCore's Proposed SAN Click here to enlarge

DataCore's SANsymphony uses an in-band architecture, placing its software on an Intel server in the Fibre Channel data path, to provide virtualization to the rest of the storage network. The drawback of in-band virtualization is the possibility of the engine becoming a choke point. But because DataCore uses its in-band method to do caching, throughput should increase under most circumstances.

DataCore understood that EDC's architecture is insufficient to allow for true failover. The company addressed this issue in its implementation of SANsymphony, which can be scaled by adding memory to the unit for caching and faster processors for additional I/O throughput. Each SDS (Storage Domain Server) has a limitation of 512 TB, but the bandwidth will run out long before a single storage node reaches that capacity. Additional Fibre Channel interfaces on the SDS can increase throughput, and you can combine up to three of these devices for load-balancing. DataCore recommends a collection of Dell x86 Intel-based servers and Gadzoox switches. The vendor also recommends several Dell NAS head units for Ethernet access and suggests that EDC attach the storage on the Sun E-10000 with SCSI-Fibre Channel bridges and SSA (Serial Storage Architecture)ıFibre Channel bridges, as opposed to direct attachment. This change would ease the configuration of EDC's legacy storage. DataCore recommends an additional TB of storage via Fibre Channelıattached ATA arrays. The vendor does not give strong opinions about where to purchase this additional equipment but estimates that it should cost about $49,000--a figure included in the overall quote.

Dancing Around Data Recovery

DataCore was hesitant in addressing the disaster recovery/business continuance part of our RFI, and with good reason. To achieve true mirroring for disaster recovery/business continuance, a vendor would need more information regarding daily storage traffic and utilization on the T1 circuits. Nonetheless, DataCore responded with the caveat that a more thorough evaluation of EDC's data rates and consumption is necessary. With that, the vendor proposed SANsymphony Snapshot software to provide mirroring, a perfectly acceptable solution.

Of the responses we received, only those from DataCore and StoreAge provide details about the capabilities of the vendors' policy managers. DataCore recommends its Network Managed Volumes for automatic resource allocation, eliminating much of the need for specific policy setting past the initial installation. DataCore's SANsymphony also provides several QoS (Quality of Service) options to ensure that critical servers get the storage bandwidth they need during peak times. These can be set based on IOPs (I/Os per second) or MBps. SANsymphony will throttle other servers to provide bandwidth for servers it has been told are critical. By comparison, the StoreAge Virtualization Manager (SVM) automatically adds capacity when the virtual volume is running low on space.

Ramp-up and implementation times are crucial to business planning. Slow and careful rollout, especially in a sensitive environment, such as a financial services firm like EDC, is essential. The plan to add a layer that will affect every production system in the data center needs to be detailed and realistic.

DataCore outlined a phased approach we liked. The proposal took into account hardware, software, disruptiveness of activity and recommendations for scheduling parts of the hardware upgrades during planned service outages. This kind of realistic plan makes VPs smile. The total time for implementation is 112 hours--not the shortest estimate, but the only one we find realistic.

Now for the unrealistic: DataCore projected an ROI of nearly $35.5 million, assuming a Gartner estimate of $2.6 million lost per hour of downtime for a financial services organization. Wisely, DataCore then recommended that a storage evaluation be done to allow for a true ROI calculation. We agree. DataCore's ROI figure is so outrageous it must be recalculated, despite DataCore's extensive justification.

Founded in 1998, DataCore employs a channel strategy to resell its products. The vendor has reached several deals with IBM and Hitachi, and therefore has solid financial ground to stand on as well as an implicit endorsement from those two companies. That's not enough to sway us, but it adds to our sense that DataCore is a company to watch.

DataCore Software Corp., (877) 780-5111, (954) 377-6000; www.datacore.com

FalconStor Software

FalconStor has a realistic implementation plan and gave competent answers to difficult questions. This proposal had the second highest price, $3.98 million, surpassed only by Compaq. That's not surprising considering FalconStor also included a substantial amount of upgrade into the data center, to provide redundant data paths and a place for its IPStor to run.

Unlike the other vendors here, FalconStor presented an all-IP implementation. The vendor's IPStor software provides storage virtualization for a network that consists of 10 dual x86 800-MHz servers of EDC's choice, two Eurologic 1-TB arrays, QLogic Fibre Channel cards and switches, and Cisco Gigabit Ethernet switches to implement the hardware portion of FalconStor's solution. On the IBM RS/6000 and Sun ES-10000, EDC's big iron, FalconStor recommends protocol-converting routers from ViTec and Crossroads Systems, respectively. This hardware sits between the storage and the target servers, including the RS/6000 and the ES-10000.

FalconStor's Proposed SAN Click here to enlarge

Because we think SANs are moving toward the all-IP transport model, FalconStor's IP-based proposal has an advantage. The vendor is making sure its product is ready when the iSCSI specification is finished so existing customers can upgrade relatively easily.

We also appreciate FalconStor's ability to add IPStor functionality to existing LUNs (logical unit numbers) to ease transition. With most other solutions, the data has to be moved off, the array virtualized and the data returned--a time-consuming process.

FalconStor projects it should take 40 to 45 days to implement its solution for EDC. The project involves installing the IPStor servers and other architectural changes to the network. FalconStor was careful to say that a complete storage assessment needs to be done to complete such planning.

FalconStor's Storage Architect team can create the policies and manage storage allocation. At the vendor's platinum support level, storage needs and practices are reviewed and revised as the storage situation changes. This kind of review is wise in a rapidly changing storage environment.

FalconStor's solution is a bit unusual because of its in-band IP basis for virtualization. The IPStor offers a virtual-drive-to-virtual-drive replication not dependent on same-same hardware. Furthermore, because FalconStor is IP-based, there's no need to use conversion bridges in the replication process to convert from Fibre Channel to IP. Standard QoS and IP security can be applied to the traffic from the FalconStor to provide continuous service. However, based on its understanding of the amount of traffic EDC will generate, FalconStor advised us to upgrade the communications between the data centers to OC-3 (155.52 Mbps) to handle the sustained data transfers.

Scalability for the IPStor application largely depends on the hardware on which it's running. If the storage increases beyond a certain point, you need to add more and/or better hardware, based on the size of the IPStor servers implemented and the amount of storage data traffic being generated. IPStor is managed from a single console, so this adds only minimally to the management burden of the solution.

FalconStor's proposal requires a huge amount of hardware that needs lots of horsepower and therefore more server maintenance and installation.

The company's ROI estimates, however, seemed sensible. Cautioning that a storage evaluation had to be done beforehand for a real ROI, FalconStor showed us a three-year projected savings in hardware and software costs of more than $300,000. Companies still need to decide for themselves what components constitute ROI, but FalconStor's approach makes sense to us.

We're concerned about the company's age--two years old--and size--110 employees. We don't think this issue is a deal breaker, but it is a consideration. Four million dollars is a lot of money to give to a company younger than some staplers in your office.

FalconStor Software, (631) 777-5188; www.falconstor.com

StoreAge Networking Technologies

StoreAge's proposal was good but not the best. We appreciated the executive summary at the beginning of its response, but otherwise, the vendor covered the bases without any particular depth. We like StoreAge's technology, and in many ways it's the easiest to understand. The company's out-of-band architecture is conceptually simple, and the solution is less expensive than the others, a paltry $188,600.

The heart of StoreAge's proposal is its SVM appliance, which has the vendor's software loaded on it. StoreAge recommends that EDC install two devices at each data center and addresses fault tolerance by suggesting the SVM appliances be installed in dual-redundant configuration. If either unit fails, the other takes over. To replace a failed SVM appliance, unplug the unit and plug in the replacement, which will synchronize with the active unit and receive all its configuration data automatically.

Because StoreAge, like Compaq, uses an out-of-band configuration, even if both appliances fail, the data will continue to pass between hosts and storage devices. There is an agent loaded on each machine that needs virtualization.

To support EDC's site-mirroring strategy, StoreAge recommends using MultiCopy, its snapshot application, to ensure the replication of data between the data centers. StoreAge did not feel that any change was needed in the communications between the two data centers for a minimum of two years; we think that estimate is a little optimistic.

StoreAge's out-of-band architecture is unaffected by changes to the SAN. Brands and types of storage do not matter. However, the solution does need additional capacity keys (licenses) to reflect the changes in the SAN, assuming changes bring the SAN beyond the size limitations of the current keys.

A few aspects of the proposal worried us, particularly in the area of implementation. The vendor's claim that its solution can be implemented in eight hours seems too optimistic. We were also disturbed that StoreAge failed to mention the inadequacy of EDC's storage infrastructure in terms of redundant reliability. We understand that StoreAge doesn't sell equipment to solve this problem, but the vendor should have recommended a hardware upgrade rather than suggesting that we consolidate the hardware to ease management. This would have given us much more confidence in the response.

Reporting, trending and capacity planning are the cream cheese on the bagel of storage networking, yet StoreAge lacked a commitment to these critical elements. Instead of providing tools to view or sort data, the vendor merely keeps a log file of the virtualization software's activities. StoreAge relies largely on third-party storage-resource-allocation packages, SNMP traps and the Fibre Channel MIB.

StoreAge didn't provide specific ROI figures but instead pointed out some its solution's benefits, such as centralized management of storage resources, reduced downtime and use of common methodologies. Much of what the company mentioned is the common soft ROI listed for virtualization and SAN vendors.

StoreAge Networking Technologies, (866) 658-2804, (949) 754-0640; www.store-age.com

Veritas Software

Veritas is a monolith in the storage and backup industries: the eighth largest software vendor in the world in terms of revenue. In terms of stability, recognition and reputation, Veritas is a go-to company. You would think this level of success would have produced a solid response to our RFI, but the company's proposal was lacking. Veritas, a former division of Seagate Technology, put forth minimal effort. Since Veritas characterized its solution as highly available (up almost all the time) rather than fault tolerant (up all the time), we can't recommend it for a financial services firm such as EDC.

Veritas suggests using the volume replication feature of its Volume Manager application to address the site-mirroring issues between the two sites. It also offers add-on clustering and global cluster manager software to facilitate data and application recovery over long distances. Veritas didn't discuss the effect its solution would have on the T1 circuits between the data centers.

The vendor ducked our questions about scalability and its ability to help EDC avoid technological change. On the scalability front, Veritas claimed we would not need to add systems administrators and reassured us that the Database Edition (with Volume Manager and File System) would make life easier for database administrators. That's fine, but it doesn't address what EDC will have to do with software as growth continues and whether EDC will need additional licensing.

They Didn't Call, They Didn't Write

Veritas obviously had some questions for us, as the vendor indicated at one point that it didn't know what we were asking. We made ourselves available for discussion in a variety of ways, and the other RFI participants took advantage of this arrangement. If this RFI response is typical, we wonder if its size and place in the storage world has gone to Veritas' head.

On the question of technological change adaptation, Veritas misinterpreted our concern and explained how it could facilitate hardware upgrades by making moving applications to an alternate machine simple. This is a benefit, but not what we sought.

We were further dissatisfied that Veritas didn't point out the inadequacy of our storage network from the standpoint of redundancy. We would expect our storage vendor-partner to tell us if we have a problem bigger than the need for virtualization, but we heard nary a peep from Veritas.

In the areas of trending and capacity planning, Veritas referred us to its experienced group of field sales strategists. We think this is an adequate response, given the information in the RFI, but would have felt better if the word sales weren't in the representatives' titles. It's like inviting an encyclopedia salesperson into your home. On the reporting front, Veritas' reports and analysis tools are more than adequate.

Like StoreAge, Veritas did not incorporate any numbers into its ROI discussion but listed advantages, such as improved performance, less downtime, less management time and less required labor. That's pretty standard ROI justification when dealing with SAN and virtualization vendors.

Veritas Software, (650) 527-8000; www.veritas.com