Starter SANs suit many needs

28.01.2005
Von Logan G.

Easy setup and enterprise features of Compellent Technologies Inc., Dell Inc./EMC Corp., DNF, Hewlett-Packard Co., IBM Corp., iQstor Networks Inc., MPC Computers LLC, and Nexsan Technologies bundles hit the mark for "es and corporate networks The small business sector is poised to become the next big market for SANs. For vendors courting this market, the biggest challenge is that because small businesss typically lack specialized storage architects and administrators, building SANs from components is a tricky proposition for them.

To see how some of the major players are addressing this issue, I looked at small business-friendly SAN offerings from Compellent, Dell/EMC, Dynamic Network Factory, Hewlett-Packard, IBM, iQstor, MPC, and Nexsan. Of the eight systems tested, three offer FC (Fibre Channel), three offer iSCSI (Internet SCSI), and two offer both forms of connectivity.

Although marketed to smaller shops, these are enterprise systems at their hearts -- with some capable of scaling to dozens or even hundreds of terabytes. Easy to set up and use, each system offers performance suitable for all but the most demanding, specialized applications. You might assume iSCSI wouldn"t offer performance on par with that provided by FC, and this would be true for iSCSI over Gigabit Ethernet as compared with a 2Gbps FC link. But iSCSI can run over 10 Gigabit Ethernet, and with the right hardware and software in place, that combination could easily outperform 4Gbps FC.

Similarly, although the systems tested use FC, SCSI, or SATA drives, a drive"s performance isn"t necessarily a good indicator of the overall performance of a SAN system. The storage processor, network interface, caching, HBA, and other components can affect performance just as much or more than drives do. In my tests of the eight systems, I found no startling differences in performance. Regardless of the type of interface or drive used, all proved more than capable of filling the bandwidth to the box.

Some people make an issue of the greater MTBF (mean time between failures) of FC and SCSI drives (1 million hours or more) when compared with those of SATA and ATA drives (400,000 hours or more). But even with a 400,000-hour MTBF, assuming an active life of five years (approximately 43,800 hours), it"s likely that a SATA drive would be replaced before it would fail, and it is capable of providing three times the capacity of an FC or SCSI drive at a quarter of the cost.

A complete FC SAN system includes HBAs, a switch or direct cable connection from server to storage, and the storage subsystem. Dell/EMC and HP provide complete kits with FC HBAs, switches, and storage arrays. These kits make the installation and configuration of a SAN simple -- no worrying about getting matching components or whether they will work together properly. Both the Dell/EMC and HP systems include easy installation and RAID management software that make it easy to get the systems running.

This is not to say the other systems are difficult to install. The days of having to input long identification strings before HBAs and storage arrays can communicate are past. All the FC systems were easily autodiscovered and configured, and the software included with each system provided easy configuration and management.

Click for larger view.Technically speaking, iSCSI systems don"t need HBAs -- the iSCSI drivers included with Windows Server 2003 and available for Linux will allow you to use any Ethernet adapter to connect to an iSCSI storage array. For optimum performance, however, you should consider an Ethernet adapter that offloads iSCSI processing, which reduces the load on the server and optimizes network communications. Likewise, you"ll get better performance running iSCSI over an Ethernet network that is separate from the company LAN.

The choice between iSCSI and FC comes down to price vs. performance. For a given capacity, there isn"t a huge price difference between iSCSI and FC storage arrays, but iSCSI (Ethernet) switches and iSCSI HBAs are much less expensive than FC switches and FC HBAs. On the other hand, 2Gbps FC is more than twice as fast as iSCSI over Gigabit Ethernet. Given the relative volumes of Ethernet vs. FC sales and the rate at which 10 Gigabit Ethernet prices are dropping, it may not be long before 10Gbps iSCSI is cheaper than 2Gbps FC -- and it will be cheaper than 4Gbps FC even sooner.

Since all the products in this review offer relatively simple installation and administration, the prime differentiator -- after you choose FC or iSCSI -- will be the storage management features available. These features include snapshots, backup, replication, remote replication, on-the-fly expansion of existing partitions and volumes, the ability to change RAID levels without disrupting stored data, and more. Whether you"ll want these features will depend on what you"re using the storage for and whether you already have storage management software.

One final issue you"ll want to consider is that disk drives for a given array are typically available only from the SAN system vendor. This is due to the need to qualify drives and ensure that they"ll operate with existing hardware. If you buy a system with less than a full load of drives, be aware that you"ll have to go back to the vendor for more drives as necessary. The exception to this is the HP system, which uses the same Ultra320 SCSI drives that are available from many retailers for use in HP servers.

CompellentStorage Center QuickStart

Compellent"s QuickStart is considerably more expensive than the other systems in this test, but it also delivers exceptional performance, ease-of-use, and features that the other systems don"t have. Compellent even includes on-site support for the initial QuickStart installation.

The Compellent system consists of two parts: the storage controller (housed in a separate 2U box) and one or more storage units. The storage units hold as many as 16 disk drives -- either FC drives or SATA drives or both, allowing you to include a high-performance database volume and a cheap file-server volume in the same unit, for example.

I received one storage controller and two storage units, one with nine Maxtor 250GB SATA drives and one with nine Seagate 147GB FC drives. The controller had four dual-port FC interfaces, and each storage unit had two dual-port FC interfaces (iSCSI is optional). All had redundant fans and power supplies.

The storage controller supports as many as three dual-connect, direct-attached storage units -- more if you use FC switches to connect additional storage units to the controller. After you have connected the storage unit to the controller and the controller to your server, the system is ready to run. After an initial network configuration via serial terminal, the rest of the configuration is done through a browser. The interface is straightforward and easy to navigate and includes useful online help. Users and groups can be configured with differing levels of access so that some administrators can be restricted to monitoring and viewing data, while others can be given the access necessary to make changes.

Administrators who are in a hurry or who aren"t sure exactly how they want to configure their storage will appreciate the Compellent system. Whereas some of the other storage arrays took a half hour or more to create a 300GB RAID 5 volume for the first time, the Compellent system created a 2TB RAID 5 volume in approximately 30 seconds. (Creating the partition and formatting the drive will take additional time.)

Click for larger view.Volumes are created using an excellent wizard-based utility, which asks the right questions and guides inexperienced admins past the usual pitfalls, such as not leaving room on a partition for snapshots. In addition, nothing is set in stone with this system because volumes can be expanded or contracted on the fly, without disrupting user access, let alone requiring a reformat of the volume.

You can even add drives to an existing volume if you have prepared for it, using what Compellent calls Dynamic Capacity. This feature allows you to create a virtual volume that exceeds the current physical capacity of the system. When you add more drives later, you can allocate them to the existing partition without interrupting access to data.

Through use of Compellent"s virtualization service, Storage Center Core, volumes can also span multiple storage units, as well as different types of drives. This means that all the drives in a volume aren"t required to be the same size or brand, as most other systems require. I also found that changing RAID levels -- from RAID 10 to RAID 5 to RAID 1 -- can be done on the fly, without disrupting user access.

An unusual feature is what Compellent calls Data Instant Replay. This allows admins to schedule regular captures of changing data files and to map the capture, or replay it, to another drive letter or even a different server. If a user accidentally deletes a file or erroneously overwrites an existing file, the administrator can point the user to a copy of the appropriate version of the file. The process is intuitive, very fast, and puts most of the recovery operation in the hands of end-users.

The Compellent QuickStart systems are expensive and represent the high end of this roundup, but they"re also the most capable, offering extensive storage management features, great performance, and incredibly fast setup of RAID volumes. The monitoring tools are also excellent, showing historical trends in data access, wait times, and more, which allows for easy diagnosis of storage problems.

Dell/EMC AX100

The Dell/EMC AX100 is a starter kit for FC SANs that comes with two QLogic HBAs, a Brocade SilkWorm 3250 eight-port FC switch, and a storage subsystem that supports dual controllers and as many as 12 SATA drives. The dual-controller configuration requires a dedicated UPS, which is also included. The UPS backs up the cache, ensuring that data is preserved even in a power outage. The configuration I received had eight 250GB drives; dual controllers; mirrored, dual redundant cache; redundant power supplies; the UPS; the Brocade switch; and all the necessary FC cable and software.

The AX100 storage subsystem is a full-featured enterprise enclosure with two dual-port FC controllers, redundant power supplies, and network connections for out-of-band setup and administration. The EMC Navisphere Storage System Initialization software automatically discovers the AX100 and performs basic configuration, while the Navisphere Server Utility allows for the easy configuration of snapshots and access to snapshots. Configuring the SilkWorm switch is similarly straightforward.

Installing the AX100 is a simple matter of installing the HBAs in the server or servers, making all the appropriate connections, and powering up the system. You then run the Initialization and Server Utility software, both of which are very straightforward. Finally, you can configure the storage array itself using a Web browser. The browser application works well and the interface is easy to navigate.

Setting up the storage system is simple enough, although there is one oddity: The subsystem OS resides on disks 0, 1, 2, and 3, so those disks already have small partitions on them. As a result, creating a disk pool that includes those four disks plus others isn"t recommended. Users who want to set up a single, large partition can do this, but the system will warn you it"s not a good idea. There are a number of places where the system will tell you if you"re doing something potentially dangerous. For instance, if you use all the disks to create a disk pool, the system will warn you that you should have a hot spare. You can always go ahead anyway, but the prompts help less experienced administrators stay out of trouble.

After you have created disk pools, you can create virtual disks, which are presented by the controller to the OS as a single attached hard drive. The difference is that you can resize a virtual disk if you have unallocated space on the AX100.

In addition to creating storage and managing it, the browser interface allows you to configure alerts, and it shows the status of the array and associated hardware, including power supplies, fans, and drives. One small irritant is that there is no power switch on the system, which can be powered off only by unplugging the unit.

The Dell/EMC AX100 is an excellent starter kit. For less than US$10,000 (starting price), you get a complete SAN that includes solid redundancy, good storage management software, and a migration path to the EMC CX enterprise-class arrays.

DNF IPBank-E 1600

A relatively small company, Dynamic Network Factory is not nearly as well-known as Dell, EMC, HP, or IBM, but it offers solid storage management features and a good value in terms of gigabytes per dollar. An iSCSI storage system running Windows Server 2003 and FalconStor"s IPStor, the IPBank-E offers snapshots; Windows Volume Shadow Copy; mirroring; clustering, using mirrored or remotely replicated IPBank appliances; disk-to-disk backup; and virtual disks for on-the-fly expansion of volumes.

The IPBank-E 1600 is a 3U server with 16 drive bays and lots of room for expansion. According to DNF, the system will support as many as eight 10Gbps iSCSI HBAs, each of which can support as much as 8Gbps of true throughput. Other adapter options include multiple 1Gbps iSCSI, FC, and InfiniBand HBAs.

The system I received for testing had 16 238GB SATA drives, two 3ware eight-drive SATA RAID controllers, and dual 1Gbps iSCSI interfaces. Setup consisted of plugging in the system, connecting it to the network, and turning it on. It acquired an IP address from the DHCP server. All I had to do was point a browser at the IP address, and I was able to configure the system.

The browser interface is clean and easy to navigate, and the necessary settings are easy to step through. Setting up virtual disks and volumes was simple and well-documented. The IPBank-E offers excellent snapshot capability, including specialized snapshot services designed to run with Windows Server 2003"s Volume Shadow Copy, Microsoft Exchange and SQL Server, and Oracle applications. Backup and restore features are also sophisticated, including disk-to-disk and disaster-recovery capabilities that make it easy to restore a storage unit from a remotely replicated counterpart.

The IPBank-E 1600 offers a good, basic set of storage management capabilities, which aren"t as sophisticated as the Compellent or MPC offerings but are easier to set up and a good deal less expensive. It is a good choice for a starter system where ultimate scalability is not a big concern.

HP StorageWorks MSA1000 " SAN Kit

As does the Dell/EMC AX100, the HP StorageWorks MSA1000 (Modular Smart Array) " SAN Kit provides everything you"ll need to set up a basic storage network, including the MSA1000, two FC HBAs, an eight-port 2Gbps FC switch, and a rack-mount kit. The MSA1000 is a dual-controller-capable, 14-drive chassis that uses standard HP Ultra320 SCSI hard drives.

The unit I received had one single-port controller and four 73GB Ultra320 SCSI drives. The overall installation is quite straightforward, from installing the FC HBAs and switch to connecting everything together. The HBA software installation is quick and easy, and the HP " SAN Wizard makes setting up the storage system simple.

Although the MSA1000 Kit seems relatively expensive -- costing approximately $12,000 for just 280GB of storage with no management software -- that price includes fast SCSI drives, two HBAs, and an eight-port FC switch. Furthermore, the system offers solid performance, as well as the benefit of using disk drives that are widely available from retailers.

The MSA1000 Kit and Dell/EMC AX100 are quite similar in terms of hardware and ease of setup. The main difference lies in software. Whereas the Dell/EMC system offers basic storage management capabilities such as snapshots and local mirroring (through the switch), the HP system requires the purchase of HP OpenView Storage Virtual Replicator and HP OpenView Storage Mirroring software to get the same features. What both offer is a complete SAN kit, including approximately $6,000 worth of FC equipment. The choice between them will depend on whether you already have third-party storage management tools on hand. Either SAN kit makes an excellent starting point, with the Dell/EMC system having a slight edge in cost and the HP MSA1000 SAN Kit providing better availability of service and parts.

IBM TotalStorage DS300

The TotalStorage DS300 is an iSCSI-only system in a 3U, 14-drive chassis with one controller, dual Ethernet ports, Ultra320 SCSI drives, and redundant power supplies and fans. The system I received included seven 73GB drives.

Setup is quite easy. The system has a default IP address, so no serial console is necessary; you just Telnet to the system for the initial network setup and iSCSI Ethernet configuration and then install the IBM ServeRAID Manager software on a workstation. The ServeRAID software allows you to configure RAID levels, snapshots, and local mirroring through a simple, straightforward interface. Volumes can be expanded without reformatting, and RAID levels can be changed without disrupting data.

The iSCSI configuration is all in one place, making it simple to attach servers. Although the DS300 does not offer all the bells and whistles that the Compellent and MPC systems do, it does offer excellent performance and easy setup at a low price: The system, with no drives, costs $3,895, and its 73GB Ultra320 SCSI drives cost $499 each.

With the proper iSCSI HBA, the performance of the DS300 should be satisfactory for most any application. Although iSCSI over Gigabit Ethernet is not as fast as 1Gbps FC, let alone 2Gbps or 4Gbps FC, it is much less expensive to deploy, and frankly, the throughput to storage will not be the bottleneck in most applications. With fast Ultra320 15,000rpm drives, the DS300"s I/O speed, which is often a bigger factor in the performance of database applications than throughput, should rival that of FC systems.

iQstoriQ1200 SATA Storage System

The iQstor iQ1200 SATA Storage System is a 3U, 15-drive subsystem with dual two-port FC controllers. As does the Dell/EMC AX100, the iQ1200 uses SATA drives to provide strong performance at a low price. The unit I received contained 15 enterprise-class Seagate Barracuda 80GB drives for a total capacity of 1.2TB. With 400GB drives, capacity is 6TB, so you have your choice of either high-performance or high-capacity drives.

You"ll also find a complete set of storage management features, including the ability to create volumes that span multiple storage units. In addition to virtual volumes and snapshots, the system supports both asynchronous and synchronous mirroring for long-distance and local replication of data, respectively.

The iQ1200 is easy to configure, requiring an initial serial console connection to set up IP address information. The rest of the configuration can be done through a browser. Monitoring status, creating RAID sets and virtual volumes, setting up local or remote mirroring, and scheduling snapshots are all simple and straightforward. If you get stuck, the documentation and online help is clear and useful, stepping you through the necessary procedures.

Priced less than $10,000 for a terabyte of storage and an excellent set of features, the iQ1200 offers a rich blend of capabilities for the buck. For shops needing storage management features, iQstor can"t match Compellent or MPC but has the edge over Dell/EMC, IBM, and even DNF.

MPC DataFrame 420

The MPC DataFrame 420 is a Linux-based, iSCSI storage array that provides two Gigabit Ethernet interfaces and 16 SATA drives by default. It can also be configured with both iSCSI and FC controllers. The system I received had 16 250GB hard drives, three SATA RAID controllers, dual FC boot drives, dual power supplies, and redundant fans.

Initial network configuration of the DataFrame 420 is accomplished via serial terminal. After the basic parameters have been set, you install the configuration application on any Windows workstation; Web-based management is not available. The Windows application can discover and manage all DataFrames on the network. Initial network configuration is well organized and easy to access. You can configure RAID levels and the RAID set, as well as network parameters such as bonding multiple Ethernet interfaces for additional throughput or redundancy.

After the basic information has been configured, you will need to set up management groups, clusters, authentication groups, and volumes. These features make it easy to automatically set up clusters that will fail over to a hot spare storage unit if one of the primary units fails, but they also complicate the setup of a single unit considerably. Most admins who want to take advantage of the advanced features of the 420 will set up the individual units as RAID 0 rather than RAID 5 and then use redundant units for data protection -- striping data across multiple units rather than within a single unit.

The fail-over system is sophisticated. A redundant unit in the cluster can assume the same virtual IP address as the failed unit so that the Microsoft iSCSI driver will continue to function. Oddly, a cluster must be â??createdâ?? even if you have only one unit. Even after the cluster has been set up, you must create an authentication group before you can configure the iSCSI server information, create a volume, associate the authentication group with the volume, and then attach to the volume.

As you can see, setting up the DataFrame 420 is a good bit more complex than configuring the other systems here. The process also takes longer than it does with the DNF and IBM systems, where all of these separate group configurations take place on a single screen. But the DataFrame ultimately offers more advanced features for automatically setting up clusters and enterprise-scale replication. It isn"t really intended to function as a single storage unit -- when it is set up, it offers additional redundancy features and easy administration of multiple units.

As does the Compellent QuickStart, the DataFrame 420 allows you to create a virtual volume that is larger than the available physical storage and then add drives to that volume without having to repartition or take existing storage offline. It also provides the same sophisticated replication features and good snapshot features at a similar price. I found the Compellent system easier to configure, but the DataFrame is still much easier to set up than the previous generation of these systems.

The MPC DataFrame 420 is a sophisticated storage system that offers both iSCSI and FC connectivity and high levels of scalability and redundancy. It shines brightest and provides the greatest return on its administrative complexity when deployed in pairs or larger groups.

Nexsan SATABlade

The Nexsan SATABlade makes for a startling contrast with some of the other systems here. At a price of $12,300, it provides 3.2TB of storage in a 50-pound, 1U pizza box. It"s even reasonably quiet. By choosing the optional bigger drives, you could squeeze 4TB into the same 1U package, a marked contrast to a 150-pound, 3U or 4U, 1.2TB system that costs more.

Setting up the unit is simple, partially because there isn"t a lot that needs to be set up. The SATABlade doesn"t include much storage management. It supports local mirroring, and that"s about it. Other features such as remote replication, snapshots, or virtualization will require third-party software. This isn"t necessarily a bad thing. For any given capacity, the savings the SATABlade offers will be more than enough to pay for storage management software. (Nexsan provides an extensive list of management software vendors that have qualified the Nexsan system for use with their products.) And if all you want is cheap, high-performance networked storage, you don"t have to pay for what you won"t use.

There"s something here for everyone, from the simple storage-only systems from HP and Nexsan to the full-featured and scalable systems from Compellent and MPC. In the middle are the systems from Dell/EMC, DNF, IBM, and iQstor. Which system is the best fit for you will come down to the types of drives supported and the storage management features you need.

The Compellent and MPC systems offer maximum scalability and functionality right out of the box. The systems from HP and Nexsan allow you to start small and build up. You can add separate storage management software to either of these systems and get approximately the same level of functionality that you"ll get with the Compellent or MPC system -- although you might end up paying more for a given capacity and capability. As does Dell/EMC, HP offers the benefit of a complete starter kit and no-brainer setup.

If you won"t need to support end-users with features such as Compellent"s Data Instant Replay or deploy a wide-area SAN as easily as MPC allows, the middle four systems all offer good levels of functionality and fault tolerance. Among them, the iQstor system offers the strongest management features, the DNF IPBank-E offers nice extras to Windows shops, and the IBM solution offers the lowest cost of entry and a great value from a tier-one vendor.