Storage Technology News:
The new breed of RAID controllers
02 Dec 2005 | SearchStorage.com
Networked storage gets all the headlines, but DAS is still a part of most companies. Deploying and maintaining it usually means picking RAID controllers. For more than a decade, that's meant buying a parallel SCSI RAID controller. But now, IT managers face some unfamiliar choices when considering what RAID controllers to use for DAS. With the arrival of SAS and SATA, and significantly faster PCI-X and PCIe busses, the game has changed. You have a range of choices, ranging in price from a few hundred dollars for simpler cards to as much as $1,000 for the most sophisticated and powerful cards.
Whether you're selecting a RAID controller for a PC or Unix server, choosing a controller for a home-grown NAS box, or evaluating the controllers in a midrange or low-end array, or building your own NAS, you have to consider several features of the card:
- Disk support
- Expansion slot requirements
- RAID features
- Port count
Server-based RAID controller issues
RAID controllers for storage arrays
"You don't call them 'RAID controllers' in a storage array," explains Ashish Nadkarni, senior consultant at GlassHouse Technologies Inc. "But they give you a much broader reach of what we can do with the controller itself with the amount of memory [cache], the performance and the features of the controller, such as snapshots."
A storage array can implement RAID for hundreds of disks using an array of RAID controllers that each provide their own level of redundancy -- when one controller fails, another will take over seamlessly. This is very different from single server-based controllers where a controller failure will disable the entire RAID system, possibly resulting in data loss.
"Storage array vendors tend to OEM from different vendors," Nadkarni says. "LSI and Adaptec are the most commonly used cards [in storage arrays] that I know."
Cache sizes are also much larger. A single controller card may only offer a few megabytes of cache, but RAID controllers used for storage arrays may offer hundreds of gigabytes. Memory allows a fast data transfer from the network, allowing the user to continue working while the data is transferred to disk later. Cache is commonly operated with a battery backup to prevent data loss in the event of a disk/controller fault.
Nadkarni explains that fan-in and fan-out ratios are extremely important in storage array planning, representing the very best and worse connection scenarios that the array will handle.
"If you want to connect 100 hosts to a single array, you had better make sure that the array can support that number of hosts," he says.
Consider the protocols supported by the RAID controller, such as Fibre Channel (FC), iSCSI, NFS, and so on. As an example, Adaptec's 7211C host bus adapter is designed for iSCSI SAN applications. Although drive selection has no direct impact on RAID protection, the choice of FC, SATA, SATA II or SAS drives will influence the cost/performance tradeoff of enterprise RAID implementation -- especially in large storage arrays.
Finally, RAID implementations will require some amount of IT management. While management is usually light for limited server-based controllers, storage arrays with many hundreds of disks may demand significant management effort. Before selecting an array RAID controller, understand how management tools and tactics may vary with your choice of controller.
Disk support, drive ports and port speed. The big change in controller requirements is the diversity of disk interconnects that you might want to utilize. As you take advantage of the cost/performance advantages of a tiered storage approach, you'll have to decide whether you want to mix and match disk types within a server. RAID controllers, such as Adaptec Inc.'s SCSI RAID 2200S ($799) offer two SCSI channels for traditional SCSI drives (up to UltraSCSI 320). However, many RAID controllers offer up to eight or more ports for SATA, SATA II and emerging SAS hard disks. One common example includes Adaptec's SATA RAID 21610SA card ($995) with 16 internal SATA drive ports. As SAS drives emerge, SAS controllers and dual-use SAS/SATA controllers are becoming more common. This includes the Adaptec 4805SAS PCIe card supporting up to eight SAS, SATA or SATA II drives. If so, even if you don't want to mix and match drive types within a cabinet, you may want to standardize on a single card that can be used for any drive type in any cabinet.
SAS controllers, like the LSI Logic Corp.'s MegaRAID 8408E, support both SAS and SATA II drives simultaneously, possibly enabling a small amount of storage tiering on the same RAID card.
Ashish Nadkarni, senior consultant at GlassHouse Technologies Inc., agrees that the practice is technically possible, but not always desirable. "Mixing and matching different drives on the same controller may not be a good idea," he says, "since the performance characteristics of each drive type may differ greatly -- causing imbalance in [overall RAID controller] performance."
As a rule, always verify support for mixed drive types with the controller's manufacturer. Check the transfer rate per port to ensure that the card can pass an adequate volume of data at each port. For example, the LSI Logic MegRAID SATA 300-8X and Adaptec SAS 4805SAS cards can handle up to 3 Gbps at a drive port.
Your next consideration is whether all your storage is going to be internal or some will be externally direct attached. Some of the new crop of controllers offers external ports, allowing drives (and drive racks) to be attached outside of the server. The new serial interconnects give you more flexibility and distance with externally attached devices.
Expansion slot requirements. RAID controllers must be inserted into an available PCI, PCI Express (PCIe) or PCI-X slot in the server itself. Consequently, the RAID card must match the slots available in a given server. PCI-X 2.0 has become a popular expansion architecture for servers, offering a 64-bit data bus operating up to 533 MHz (resulting in a maximum theoretical throughput of 4.3 GBps, compared to 132 MBps transfers for conventional 32-bit PCI slots at 33MHz).
PCI-X products are backward compatible with conventional PCI slots. This allows newer PCI-X controllers to reside on older servers, though the controller's performance is limited by the older expansion slot until the server can be upgraded. PCIe is an alternative high-performance expansion bus technology using between one and 32 serial data lines (dubbed "lanes") to carry data at bandwidth rated to 16 GBps.
Today, you'll find RAID cards available for all three expansion bus architectures.
RAID levels and total array size. RAID controllers are also differentiated by the variety of RAID implementations (or "RAID levels") they support. Most RAID controller cards will support common RAID levels including RAID-0, 1, 5, 10 (1+0) and 50 (5+0).
More current controllers will handle the emerging RAID-6 and 60 (6+0) levels intended to protect against dual/simultaneous drive failures and is strongly encouraged when large arrays must be protected. Cards like Adaptec's SAS RAID 4805SAS ($995) card will support RAID-6/6+0 as an optional feature.
The overall size of a RAID array should also be considered against storage needs. LSI Logic's MegaRAID SAS 8408E card supports RAID arrays to 2 terabytes (TB), which is typical of current SAS and SATA offerings.
Onboard cache and battery. Cache is typically used to hold data from the server until it can be written to the respective RAID drives. Controllers with 128 MB of cache are common. Both Adaptec's SAS RAID 4805SAS card and LSI Logic's MegaRAID SATA 300-8X card ($520) include 128 MB of ECC-protected DDR2 memory. By comparison, LSI Logic's MegaRAID SAS 8408E card provides 256 MB of DDR2 memory for cache.
Cache does not require a battery backup, but unwritten cache data will be lost in the event of a power failure. For best data protection, battery backup modules are typically available as an option and are often purchased with RAID controllers. Battery backups can work from 24 to 72 hours depending on the card and backup module selected.
Consider enhanced controller features
In addition to the four key issues of slot, RAID level, ports/drives and cache, there are other powerful RAID features being routinely integrated into current RAID controllers. Hot spare support (dubbed RAID-5EE) allows a fully functional drive to reside in an existing RAID group, but is left unwritten. When a drive fault occurs, that drive's data is automatically reconstructed on the hot spare. The faulty drive can then be replaced later, becoming the new hot spare. Hot swap capability minimizes maintenance and repair downtime by allowing faulty drive replacement without powering down the server or storage array.
RAID controllers are also increasingly scalable and multiple cards can sometimes be installed in tandem for additional capacity, performance and redundancy. For example, the 3ware (AMCC) 8000 family of RAID controllers can scale to 12 TB of capacity by using 12 drives at 250 GB each across four RAID controllers in the system.
Multiple controllers can support tiering if desired by dedicating a particular drive type to each controller. Although it is possible to install multiple RAID controllers in a single server (Broadcom Corp. calls this "controller spanning"), the choice to do so depends on your storage objectives.
"Sometimes [a server's expansion] slot capacity may dictate that only one card be used," Nadkarni says. "But in general using two cards instead of one will help from a performance and redundancy perspective."
Don't forget management software
Analysts note that management tools play an important role in RAID implementations, allowing storage managers to configure, monitor and manage RAID systems locally (even remotely over the enterprise network). It's common to find RAID management tools bundled with the RAID controller at the controller level and operating system level. For example, LSI Logic bundles four tools with its MegaRAID SATA 300-8X card. The MegaRAID BIOS Configuration Utility and WebBIOS allow local and remote hardware-based configuration, the MegaManager provides a command-line interface for RAID management and the Power Console Plus supplies a GUI-based monitoring and management interface.
Management tools also support increasingly powerful features appearing in new RAID controllers, making RAID arrays far more dynamic and versatile. A move toward online capacity expansion lets users grow their RAID groups as needed without a disruptive upgrade and recreation of the entire RAID group.
Native command queuing is found in SATA II controllers, allowing higher performance by reordering disk commands more efficiently. RAID level migration is another notable feature that allows RAID protection to be changed on the fly -- again without the need to disrupt data in the current group.
Brian Garrett, analyst at Enterprise Strategy Group, notes that "background RAID initialization" reduces the time-consuming process of preparing the parity for RAID-5 and RAID-6. "Historically, you've always had to wait for that [zeroing] operation to happen, he says. Vendors have gotten clever, and they'll do it in the background as a low-priority task -- giving you access to the system right away." However, disks can also be zeroed out at the factory to reduce RAID configuration times.
RAID controller manufacturers (alphabetical)
Adaptec, Inc. [Server and array cards]
Applied Micro Circuits Corporation (acquired 3Ware) [Server cards]
Broadcom Corporation (acquired RAIDcore) [Server cards]
Emulex [Array cards]
EqualLogic Inc. [Array cards]
Intel Corporation [Server cards]
Intelligent Computer Peripherals (ICP Vortex owned by Adaptec) [Server cards]
LSI Logic Corporation [Server cards]
Promise Technology Inc. [Server cards]
Looking to the future of RAID
Today, RAID requires hands-on management to organize disks, migrate RAID levels, handle spare drives, perform snapshots and so on. But analysts agree that traditional RAID technologies are likely to change as storage assumes a more ubiquitous role in the enterprise -- requiring less conscious thought and attention.
"I believe that RAID will mature into something that is no longer called RAID," Nadkarni says. He explains that the basic storage virtualization offered by RAID is systematically converging with the idea of storage as a corporate resource. The protections offered by RAID will become automatic, adapting to suit the unique needs of each business. "It may dynamically change its nature based on the performance and capacity requirements of the server." Eventually, Nadkarni expects that RAID will be a self-managing feature of the storage environment, probably tied to storage provisioning activities.
In the nearer term, analysts expect SAS drives to assume a larger role in RAID applications.
"The SAS ecosystem is just coming alive," Garrett says. "There are SAS RAID controllers and SAS drives available, but its adoption hasn't taken off yet."
He expects to see SCSI RAID features on the motherboard eventually shift to SAS as a natural evolution. Over the longer term, analysts predict additional intelligence at the hardware level.
"The electronics designed in a [RAID] controller will become smaller and eventually make its way into the disk," Nadkarni says. "The controllers themselves may become part of a fully virtualized multiprotocol environment where you won't talk about controllers or arrays. You'll just talk about a 'resource.' "