February 1998

RAID GETS FASTER AND SMALLER

RAID is faster than optical jukeboxes and tape libraries. When you need to get data fast, you need RAID. Today's RAID systems use faster drives. Some arrays are small enough to fit on your desktop.

There have been several advances in RAID over the last year. These include:

1. Faster drives. New 10,000 RPM drives improve data transfer and I/O rates. You can access the data on the disk more quickly because the rotational speed is faster. RAID arrays with 10,000 RPM drives improve performance by up to 37% over the same systems with 7,200 RPM drives.

This speed comes with a bit of pain. The 10,000 RPM drives require more power and generate a little more heat and vibration. The RAID cabinets have to be able to handle that. Some have extra fans to prevent overheating and rails that prevent the drives in the array from spinning out of control.

2. Smaller RAID arrays. A new flock of small desktop RAID systems let individuals have their own RAID for storage intensive work like video editing.

"A few years ago there was no demand for something this compact," says David Snook, product line manager at Storage Dimensions (Milpitas, CA 408-954-0710). "You didn't put 45 gigs of storage on your desktop -- there was no need for it."

3. More fibre channel. Many RAID systems have some fibre channel support. Fibre channel with its 100-megabytes-a-second data transfer rate eliminates SCSI bottlenecks.

For video and multimedia applications there's no alternative to the bandwidth fibre channel delivers. Fibre channel lets you add many RAID devices to your system -- up to 126 on a loop. SCSI is limited to eight to 15 devices on an individual chain.

Host computers have begun to support fibre channel. "We've seen a huge migration on the part of host manufacturers to fibre channel," says Paul Danahy, product marketing manager at Clariion (Southboro, MA 508-480-7280). "Hewlett-Packard, Silicon Graphics, Sun have all announced fibre channel products. IBM also has fibre channel drives on the market."

4. Network RAID. There are now RAID products that attach directly to a network without being connected to a host computer. This lets people on a network access shared files in a centralized RAID storage system faster than if they had to go through a server.

RAIDing the Desktop

Desktop RAID gives individuals and small workgroups the benefits of RAID storage in a small footprint. Imagine 50 gigabytes of images and data in a small box on your desktop, instantly accessible. (Also imagine having several thousand dollars less in your bank account.)

On the desktop, as with standalone towers and cabinets, RAID tasks can be controlled with hardware or software. Software RAID is less expensive than hardware RAID. You can use whatever drives you want. An example of software RAID is Windows NT. This has support for RAID levels 1 and 5 built into it.

Like other types of software RAID, NT's RAID sits on your CPU, using processor resources. If you don't use your computer for much besides RAID storage, that may be fine. Software RAID can handle only limited amounts of data, usually up to six gigabytes. And although software RAID controllers are usually cheaper than hardware RAID controllers, most of RAID's cost is in the drives. Using NT's RAID solution and buying drives separately may not be much cheaper than buying a RAID system with its own hardware RAID controller.

Hardware RAID is more heavy-duty. It can be used for networks and high-volume applications. Some hardware RAID solutions put the controller card in the host computer. Others house the controller in the RAID chassis. The RAID chassis can be inside or outside the computer.

Some Compaq and Hewlett-Packard workstations have internal hardware RAID built in. This can work well. You get automatic host/RAID agreement. Internal RAID also has limitations. You can only have as many drives in your RAID system as you have bays in your computer.

External RAID -- a box of RAID drives outside your CPU box -- lets you expand as far as the RAID system will accommodate. You can keep your storage when you change servers. It lets you hot-swap drives.

Here are some external desktop RAID products that manufacturers have introduced in the last year:

Aiwa's (Irvine, CA 714-862-0200) new MicroArray III ($4,500) has three 5.1 gigabyte drives in a box that fits into a standard 5 1/4" drive bay. It provides 10 gigabytes of storage on its own, 100 gigabytes if you daisy-chain several together. Ultra SCSI provides a data transfer rate of up to 40 megabytes a second. Aiwa's ArrayView monitoring software provides audible and visible alarms that tell you when there's a potential malfunction.

The UltraDaytona RAIDarray ($6,000+) from CMD Technology (Irvine, CA 714-454-0800) with an expansion chassis provides up to 117 gigabytes of storage. In some tests it has achieved over 37 megabytes per second data transfer rates. The CMD StorageView Lite software that comes with the hardware lets administrators see the status of all RAID components, create hot spares and rebuild RAID sets.

Micronet's (Irvine, CA 714-453-6000) Advantage RAID puts 20-30 gigabytes of storage on the desktop for $7,000+. It supports RAIDs 0, 1 and 5 and connects to an Ultra Wide SCSI port or SCSI host adapter. It offers a 32-bit RISC microprocessor, two megabytes of RAM, battery-backed cache, redundant power supplies and an LCD panel in the front to monitor drive activity.

Promise Technology's (San Jose, CA 408-452-0948) iRAID systems let you use enhanced IDE drives in their arrays rather than SCSI drives, yet connect the array to the controller and to the host via SCSI-2. IDE drives are less expensive than SCSI drives, making the overall cost of these solutions lower than usual -- $2,000 for iRAID5, which holds up to 16 gigabytes of data, and $4,000 for iRAID10, which stores up to 32 gigabytes per array. Both RAID systems support only RAID 5.

Storage Dimensions (Milpitas, CA 408-954-0710) reintroduced its $5,000 to $14,000 RAIDPro storage systems with 9.1 gigabyte 7200 RPM low profile disk drives. This puts 45.5 gigabytes of storage in a box the size of a legal pad six inches thick.

It's priced to compete with internal storage from companies like Compaq. The system works with workstation and server versions of Novell, Netware and Windows NT. It offers Dynamic Growth and Reconfiguration. You can add a disk online without having to back up or shut the system down. You can change the RAID level as well.

Network RAID

Network-attached RAID systems give everybody on the network better access to RAID-stored data. There aren't many RAID systems that attach directly to a network yet, but many companies are working on it.

The UDSS Disk Array from Legacy Storage Systems (Markham, ON 905-475-1077) is designed for NT enterprise networking. Its direct network attach capability lets this $2,500 system provide RAID storage and fail-over for up to 32 NT or NetWare file servers on a network.

This system has a passive backplane. It holds up to 2.37 terabytes of data. It works with a wide range of operating systems including Windows 95 and NT, OS/2, Mac, Sun and Unix. Its audible alarm warns you of mechanical failures or high temperatures in the system.

Symbios Logic's (Fort Collins, CO 970-223-5100) MetaStor family of scalable storage solutions are server- and network-attached. The product line is actually modular building blocks. The DM1000 drive chassis holds up to 10 drives and the CM1000 command module contains RAID controllers. Together they scale from workgroup storage (8-90 GB) to departments (40-450 GB) to data centers (160 GB to multi-terabytes).

Workgroup configurations are $29,000 for the server-attached system and $41,000 for the network-attached version. A department configuration is $160,000 for the server-attached, $170,000 for the network-attached. Data center configurations would be about $300,000 for the server-attached and $250,000 for the network-attached.

Why use lots of little building blocks instead of one big RAID system? "First, there's the flexibility of being able to implement your enterprise how you want to," says Jim Griffin, product marketing manager. "You don't have to offer a huge centralized monolithic storage environment. In fact, most people don't.

"You might have a workgroup with certain requirements and departments with other requirements. You might have a remote organization that has other requirements or a very large data center. Or you may only be able to afford 300 gigabytes of storage now, but know that your budget and needs will increase in six months time.

The server-attached modules support RAIDs 0, 1, 3 and 5. The network-attached modules support only RAID 5.

Big RAID

Enterprises, production imaging environments and large host computers usually call for giant RAID systems. These are some things to look for in large-scale RAID:

Throughput. How fast can you get your images on and off the system? Both data transfer rates and I/O rates are clues to the system's performance. Faster drives in the RAID array give you fast I/O rates. The fastest electromagnetic drives are 10,000 RPM. Solid state disks, the fastest semiconductor drives, offer instant access to data. They're also many times more expensive than electromagnetic drives.

Throughput scales with size. The more drives and connections to the host a system has, the faster the system's potential throughput can be.

Capacity. Don't just look at raw capacity. Determine the amount of storage the system will provide with the level of RAID you plan to use. If RAID 5 suits you and you store different file sizes, look for the RAID 5 capacity of the storage system. This is 20% lower than raw capacity. RAID 1 storage comes to about 50% of total storage of a system.

Expandability. Your storage needs will grow. Check out how easy it is to add new drives and cabinets to the system. Ask if you can add new drives without having to stop the system, back up and reconfigure. Find out the system's total capacity limitations.

Fibre channel arbitrated loop RAID products can link up to 126 RAID arrays in one system. You don't want more than five devices on a SCSI chain.

Redundancy. You need at least two of everything -- controllers, power sources, fans, etc. If one conks out, the other takes over.

Monitoring. All RAID systems tell you when a drive or component fails or shows signs of weakness. Proactive methods like beepers and flashing red lights are better than systems that make you log in to a Web site or administrative program to find out what's happening.

Fibre channel support. A SCSI connection (with a maximum 40 megabytes/second bandwidth) to your host can be a bottleneck to RAID I/Os and data transfers. A fibre channel connection to your host provides bandwidth of up to 100 megabytes a second per channel. Fibre channel lets you use many RAID arrays together. An Ultra SCSI bus can't productively handle more than five RAID devices. A Fibre Channel Arbitrated Loop can link 126 RAID arrays. Fibre channel host adapters cost several hundred dollars more than SCSI host adapters.

Data warehouses, Internet servers, email, multimedia, video and transaction processing applications could all benefit from the higher bandwidth of fibre channel.

High mean time between interruptions. Figure this out by calculating the Mean Time Between Failure (MTBF) of all the RAID components, including the controller, drives, power sources and fans.

No single point of failure. No one component should have the power to bring the entire RAID system down.

Hot-swappable drives, controllers, power sources, fans. These let you pull out a dead component and put a fresh one in without stopping operations.

Hot-spare drives, controllers, power supplies, fans. Hot spares run online with the others. If a component in use fails, the spare takes over the work.

Choice of disk drives. Solid state disks or very fast magnetic drives may be best for your application. A RAID array that offers choice in drives gives you flexibility.

Mirrored cache. As you write data to a RAID system, it's temporarily stored in cache. If you have two controllers, you can mirror the cache in one on the other. If a controller fails, the data in its cache can be preserved.

Battery-backed cache. This protects your cache when the power fails.

Some of the newest large RAID products include:

Box Hill's (New York, NY 212-989-4455) RAID Box 5300 Turbo uses the company's 23 gigabyte disk drive, giving the RAID array capacity up to 1.3 terabytes.

Clariion (Westboro, MA 508-480-7280) FC5000 Series ($228,700+) of RAID arrays provide up to a terabyte of storage and are focused on data availability. They offer dual controllers, mirrored write cache and application transparent failover.

"If one pathway fails or a host bus adapter or a cable breaks, we have a redundant pathway going to the array," says Paul Danahy, product marketing manager. "The end user wouldn't see any stoppage of data availability so the application will continue running."

With two storage processors and two fibre channel loops, the FC5000 RAID arrays achieve 15,000 I/Os per second. They're designed for data warehousing, multimedia and online transaction processing. The arrays work with uniform hosts.

Clariion's new Telestor array is targeted at telephone central offices, where strict standards make sure the array is resilient in the case of fire damage, earthquake and other natural disasters. It offers up to 100 gigabytes of storage.

Digital Equipment (Maynard, MA 978-493-5111) offers a range of StorageWorks RAID arrays. The $55,000+ Enterprise Storage Array 10000 high-end solution sports a bandwidth of 56-112 Mbytes\sec and an I/O rate of 24,000 I/Os a second using two controllers. It holds 648 gigabytes on its own and multiple terabytes when several are chained together.

The RAID Array 7000 system ($41,000+) is for departments and workgroups. It holds 218 GB per cabinet. The $8,000 RAID Array 3000 is for NT servers in workgroups that need 128 or fewer gigabytes of storage. The $2,400+ RAID Array 230/Plus is used in low-end non-clustered environments. It holds up to 63 gigabytes per array.

The 7000 and 10000 RAID arrays let a mix of up to eight IBM, Sun, DEC and NT servers share access to centralized storage through a SCSI hub. All except the 230/Plus support mirrored write-back cache.

Digital says the 7000 and 10000 RAID subsystems are easy to upgrade. Users just swap out their existing Ultra SCSI boards and swap in fibre channel controllers and cache modules.

Hitachi Data Systems' (Santa Clara, CA 408-970-4468) HDS 5700E Intelligent Storage System ($29,000-$264,000) puts up to 43.5 gigabytes of storage in a RAID array that supports RAIDs 0, 1, 5 and 0+1. It comes with Conley SafePath, server software that provides automatic fail-over for server-to-RAID I/O connection failures. SafePath lets system administrators add redundant SCSI connections between data servers and RAID subsystems. For remote monitoring, the 5700E comes with SOMMET, software that lets you manage RAID storage from a network console.

Maxstrat's (Milpitas, CA 408-383-1600) Gen5-S family of RAID systems ($88,000+) achieve data transfer rates of more than 200 megabytes per second using up to eight Fast Wide or Ultra Wide SCSI host connections or four fibre channel connections. They offer over 800 gigabytes of RAID 5 storage. They can work concurrently with different computer platforms, including Silicon Graphics, IRIX, Windows NT, Sun, Solaris, IBM, AIX and HP-UX.

Micronet's $24,000+ DataDock 7000 has triple redundant power supplies and triple redundant fans. When a drive fails, this RAID system automatically spins down that drive and takes it offline. Audible and visible alarms alert users to controller, power supply, fan, over-heating and drive failures. The DataDock 7000 works with differential SCSI for distances of more than 10 feet and single-ended SCSI host interfaces. It supports RAID levels 0, 1, 5 and 3. It can run 10,000 RPM drives. Many DataDock 7000s are used for prepress workflow on Windows NT and Macintosh servers.

MTI Technology's (Anaheim, CA 714-970-0300) new Gladiator 3200 ESSRAID Array ($470,000) stores 582 gigabytes of data with one gigabyte of cache. It has active/active redundant controllers. That combines redundancy with the performance of dual I/O data paths.

nStor's (Lake Mary, FL 407-829-3500) CR8e Ultra SCSI RAID subsystem ($8,400+) works with a PCI host-based controller or a SCSI to SCSI subsystem-based controller. One enclosure holds up to 72 gigabytes.

This box is designed for users at the crossroads -- those who aren't sure whether to go to SCSI, SCSI II, Ultra SCSI, LVDS or fibre channel. It supports them all and can transition from one to another.

nStor's two new RAID enclosures, CR8f and CR18f, are based on Fibre Channel Arbitrated Loop. They provide data transfer rates of up to 200 Mbytes\sec. They support up to 126 devices in a single loop.

"Fibre is a perfect alternative because it has adequate bandwidth to support imaging and you don't have to worry about multiple buses," says Steve Paulhus, marketing director at nStor.

The CR8f, an eight-bay enclosure, is available in a tower or rack-mount configuration. It supports 4 GB, 9 GB or 18 GB fibre channel drives and provides 144 GB of storage in a single enclosure. Up to 10 CR8f enclosures can be housed in a 72" cabinet. The CR18f holds 18 1.6" drives per enclosure. These enclosures can be stacked 10 to a tower.

The nStor systems have hot-swappable power sources and drives. They support dual independent power sources. The fibre backplanes are completely passive -- there's nothing that can fail. The systems are NT cluster-ready. As soon as NT supports fibre clusters, nStor will support it. nStor offers dynamic capacity expansion -- you can add capacity to your RAID volume without taking the system down.

The Superflex 3000 from Storage Dimensions ($10,000) can provide up to a third of a terabyte of RAID 5 storage. It uses a PCI controller and needs to work with a host adapter.

The company's $9,000+ Superflex 4000 and Superflex 5000 RAID systems use SCSI-to-SCSI external controllers that connect to the server with a SCSI bus. To the server, these RAID systems look like one huge SCSI disk drive. They're host independent and can support as many as 16 separate disk drives.

Storage Computer (Nashua, NH 603-880-3005) offers their OmniRAID Cluster Array for NT ($50,000-$250,0000). "We're different because the OmniRAID can handle a mixture of applications with very rich data -- images and text," says Greg Scorziello, SVP of worldwide sales and marketing. "Traditional storage devices are good for one or the other."

The RAID system's Omniforce data migration software lets you transfer files between servers running different operating systems. One department might edit a film on an Apple Mac and then transfer the data to another department that uses Silicon Graphics computers.

OmniForce allows the migration of that file and the connectivity of that information to multiple servers. In the same building you could have an SGI, an Apple and an AS/400. OmniRAID doesn't require host-based drivers, so to upgrade or change servers you don't need a special driver.

Up to 24 NT servers can be hooked up to OmniRAID. NT clustering today only lets you cluster up to four servers. In a year or so you'll be able to cluster more.

This system doesn't require each drive to be tied to a RAID level. Every drive can handle whatever RAID level makes sense for a particular file. "We don't have multiple RAID levels," Scorziello says. "We have a storage infrastructure that spreads data across drives as a typical RAID device would do. The system can operate as if it were a RAID 3 or RAID 5.

"If you had RAID 3 you'd lock all the drives parallel and you'd transfer data in parallel. In RAID 5 you spread your parity across the drives and you transfer data in small block sizes. It comes down to architecture -- we don't daisy chain drives behind one another, every drive in our device has its own controller so it can either operate independently or in parallel.

"You might have a large file come in that has a long block size," he says. "Our hardware allows the drives to operate in parallel or asynchronously. An operating system can say, ýthat's a large block size, please transfer all the drives in parallel.'

"If another transaction comes in with a short block size, the operating system would say, ýWe can service that off one track, we don't need to transfer it off eight tracks.' It acts as if it were many RAIDs, but in fact we spread all the data in the device across all the drives."

Storage Computer invented RAID 7. "We've moved away from that numbering scheme," says Scorziello. "People say: ýWhat is it? Is it 2+5 or 3+4?' It's an intelligent operating system that we called RAID 7 to make it different. We spread data across many drives to get redundancy like everyone else does. Our architecture is parallel. Our software takes advantage of that parallelism by reacting differently to different transactions. RAID 7 is still part of our core operating system."

The Magnitude ($60,000 to $100,000) from Xiotech (Eden Prairie, MN 612-828-5980) is brand new. It boasts an astonishing I/O rate: 50,000 I/Os per second. Each Magnitude array can hold 64 drives and be attached to up to eight servers at once. "We parallelize the operation of the I/O streams in order to apply the resources to all the servers," says Greg Mangold, marketing VP.

"You could daisy-chain several smaller RAID systems together and get the same capacity levels we provide in one cabinet," he says. "But we have a series of techniques under an umbrella called REDIdata. One technique is Intelligent Virtual Disks. We virtualize up to 64 disks into one large volume that can be shared amongst all the servers connected to it."

The Magnitude connects by fibre channel to the servers. Each server gets a PCI host adapter board that connects to a duplicate board on the 600 MIPS processor board in the Magnitude chassis. The Magnitude does the majority of the data transfer and other RAID work. The Magnitude lets you mix and match drive shapes and sizes within the cabinet. It supports any operating system. Xiotech developed and tested drivers for Window NT and Novell Netware. They've scheduled drivers for Unix. The Magnitude supports RAID levels 0, 1 and 5. It can quickly grow from 50 gigabytes to a terabyte.