I chose to allocate 9 disks to one LUN and make the 10th disk a hot spare. From the Create/Delete menu, I created logical volumes of a fixed size, using a portion of the LUN's capacity. Increasing the size of a logical volume is a quick two-step process. First, you create a new logical volume segment from the same screen you used to create the primary volume. Second, you use the Configuration menu's Segments tab to add the segment to the volume you want to expand. The new space is immediately available. The NetFORCE 1500 doesn't provide a means to directly shrink a volume. To make the volume accessible to Windows users, you use the SMB/CIFS menu to create a share for the volume. File-level and directory-level permissions are easy to set with the familiar Windows Explorer Security tab on the Properties page.
Similarly, I used the NFS menu to export the volume for access by UNIX users. I used the Solaris workstation to mount the export. As I expected, I was able to access files on the NetFORCE 1500.
RAIDserver V. The RAIDserver V's Web-based configuration and administration interface is well organized and easy to navigate. From the Volume menu, you can create RAID 0, RAID 1, RAID 5, and RAID 10 arrays by selecting a check box next to each of the unassigned disk drives that you want to use for a new RAID volume and using drop-down boxes to select RAID level, segment size, and hot-spare usage. For my testing, I created a RAID 5 array—or a volume, in Raidtec's terminology. I set 1 disk to hot-spare status and allocated the other 11 disks to the array. To achieve the best performance, I set the volume to Asynchronous write mode. This mode, in which RAIDserver V signals that the write-to-disk operation is complete before the actual write to disk, enables write caching. Because of the potential for data loss (in the event of a failure that precedes the write to disk), Raidtec recommends asynchronous write mode only when a UPS system safeguards the RAIDserver V.
With the help of the Getting Started guide, I moved on to the next step: creating a share. After I completed the necessary steps, I was able to see and use a share that I created from another NT computer on the same network segment. When I added the RAIDserver V to the DNS server, I could map to and use the share from other network segments.
The RAIDserver V also supports NFS access—specifically, NFS 2.0 over UDP. Configuring NFS access requires two steps: creating an export (i.e., designating a directory on a volume as accessible to NFS users) and enabling NFS. Although the guide directed me to first enable NFS, the GUI required that I first create the export. Except for that inconsistency, the creation of the export proceeded as advertised. However, I had difficulty mounting the export. Raidtec's technical support explained the difficulty as an NFS version problem: The Solaris workstation supported NFS 3.0, whereas the RAIDserver V supported NFS 2.0.
You can make a directory accessible to both Windows and NFS users—with some limitations. When you use Windows domain authentication (rather than a local accounts database), files that NFS users create are accessible to Windows users only in read-only mode. This limitation is a result of the underlying user ID that controls access permissions. If you use a local accounts database for Windows authentication, you can create accounts with specific user IDs to match UNIX-based users' IDs. Windows domain authentication doesn't offer this functionality. The GUI doesn't let you specify UNIX-style permissions for the directory shared through an NFS export. If you use Windows domain authentication mode, the RAIDserver V knows what the user ID is after a domain user connects to a share. You can edit a domain account's user ID after the RAIDserver V knows the account.
I discovered a limitation of the RAIDserver V: It supports a maximum file size of 2GB. My test suite included random reads from a 5GB file—a test that I couldn't perform on this NAS device. The reason for the limitation is the underlying Linux 16-bit file system. To eliminate this limitation, Raidtec plans to implement a journaling file system in late 2001.
Backups and Data Snapshots
Data snapshots, which let you almost instantly capture the state of data storage from a specific point in time, are a compelling feature of four of these five NAS units. Snapshots are a relatively low-overhead method to maintain a point-in-time backup of a NAS system's files. When you take a snapshot, the NAS system copies no data. Instead, the NAS system maintains a copy of the original version of the data, often by simply saving the original block of data on disk. When you subsequently modify files, the system writes the new data to a new disk location. Therefore, a snapshot occupies only as much disk space as the original version of the changed data. Data snapshots offer a convenient way to restore files that users inadvertently delete or damage, and they let you back up volumes to tape from a specific time, while file-system activity continues.
TaskSmart N2400. Snapshot Manager, Compaq's tool for capturing a point-in-time state of a virtual disk for backup purposes, is the same tool that you use during system configuration to create storage pools and virtual disks. Snapshots appear as separate volumes that you can make available to Windows users or Help desk staff. Because the TaskSmart N2400 is a Win2K ASbased system, any enterprise backup system that works with Win2K AS will work with the N2400.
MetaStor N4300. The MetaStor N4300's Backup and Recovery sphere includes snapshot support and—if the N4300 includes an integrated backup device—a means with which to launch the backup utility. You can create snapshots manually or on a timetable that you configure. By default, the MetaStor N4300 reserves 15 percent of the file-system space for storage of the original version of the data that changed after the snapshot was taken. You can change that percentage when you schedule a snapshot. When you create a manual snapshot, you can choose a directory to place it in. The MetaStor N4300 deletes snapshots whenever you reboot the system. If you need to retain a point-in-time snapshot, you can back it up to an attached backup device.
NetApp F840. The NetApp F840 supports snapshots. As Figure 6 shows, you simply select the hours on the clock representation and designate the number of hourly, nightly, and weekly snapshots you want the system to keep. The NetApp F840 supports Network Data Management Protocol (NDMP), a standard for communication between network-based backup applications and servers. The protocol lets a tape backup device attached to the unit or to a network workstation back up data on the unit.
NetFORCE 1500. The NetFORCE 1500 also supports snapshots. (Procom Technology calls them checkpoints.) From the Checkpoint item on the File/Volume menu, I was able to create both scheduled and manual checkpoints. All checkpoints appear as a new volume; for example, checkpoints for the volume I named GB50 appear as directories in the gb50.chkpnt volume. I created a CIFS share for that volume to expose the checkpoints in a way that Windows users would find easy to access. For full data backup, the NetFORCE 1500 lets you connect SCSI tape libraries and supports NDMP client/server systems over standard Ethernet or Gigabit Ethernet links.
RAIDserver V. The RAIDserver V doesn't support point-in-time data snapshots. However, the unit offers integrated support for several SCSI-connected tape-backup devices. Although the device I tested contained no tape device, the backup options in the RAIDserver V's administrative utility seem complete and easy to use. The system supports full, incremental, differential, and custom backups—on demand or according to a schedule.
Performance Comparisons
To assess the file-server performance of each unit, I used Quest Software's Benchmark Factory to create a set of file-access transactions, then varied the simulated user load, noting the point of maximum throughput. To create a meaningful comparison, I ran the same workload against a Compaq ProLiant DL580 server with four 700MHz Pentium III Xeon processors with 1MB of Level 2 cache and thirteen 18GB Ultra 3 SCSI 10,000rpm disks. I configured this reference system with a disk and memory complement as close to that of each of the NAS units as possible. In other words, I reconfigured some or all of the 13 disks to closely match the RAID configuration of the NAS device to which I was comparing the reference system. Similarly, I used the Maxmem option in boot.ini to limit the Compaq system's available memory to the amount with which the corresponding NAS device was configured. Therefore, you should compare these test results only to the similarly configured reference system. Unless otherwise noted, I generated the test load using 47 client systems allocated over three 100Base-T network segments, with the test system multihomed to the same three segments and to a network segment hosting DCs.
TaskSmart N2400. The TaskSmart N2400 that I tested contained 1TB of disk storage in four cabinets. However, I chose to test a more modest configuration that I could match to my reference system—one RAID 5 array of 13 disk drives in one cabinet. I created a pool over the entire array and a 100GB volume within the pool. The dual-processor N2400 posted a maximum throughput of 107.7 transactions per second (tps), compared with a 136.7 tps maximum throughput for a comparable configuration on the quad-processor reference system. The higher performance of the reference system is due primarily to two factors. First, although my tests involved Ultra 3 disks in both units, the reference system's SmartArray 5300 controller is a 160MBps Ultra 3 controller, whereas the TaskSmart N2400's Smart Array 4200 controller is an 80MBps Ultra 2 controller. (By the time you read this article, another version of the TaskSmart N2400 might be available with the newer Smart Array 5300 controller.) Second, the reference system contains four Xeon processors with 1MB of Level 2 cache, compared with the TaskSmart N2400's two Pentium III processors with 256KB of Level 2 cache. The TaskSmart N2400's impressive performance comes with a hefty price tag.
MetaStor N4300. The MetaStor N4300's 67GB test volume was striped across three 6-disk RAID 5 arrays, a configuration I couldn't precisely duplicate on my 13-disk reference system. Nevertheless, testing the MetaStor N4300 produced a lackluster 85.4tps maximum throughput. I called LSI Logic Storage Systems and implemented some tuning changes, but when I retested the unit, it didn't produce significantly different results.
NetApp F840. The NetApp F840, with a RAID 4 array over seven 36GB disks and 3GB of total system memory, produced a stellar maximum throughput of 341.4tps. Network Appliance also supplied Gigabit NICs that let me test another network configuration. In the second configuration, I connected my three segments of load-generating client computers to a Foundry Networks FWS24 Gigabit/100MB Ethernet switch. I connected the test server to one of the switch's two SX Gigabit Fiber ports. The Gigabit network modestly improved the NetApp F840's overall peak throughput to 356.7tps. By way of comparison, the reference system's peak throughput improved from 163.5tps with the 100MBps network to 171.3tps with the Gigabit network.
NetFORCE 1500. I configured the 512MB NetFORCE 1500 with a 50GB volume on a nine-disk RAID 5 array. Maximum throughput with my test suite was 95tps. The similarly configured four-processor reference system posted a throughput of 87.4tps. Considering the NetFORCE 1500's cost, the unit's test results are impressive.
RAIDserver V. I couldn't run my test suite against the RAIDserver V. First, two of my test transactions read blocks of data from a 5GB file—but the RAIDserver V supports a maximum file size of only 2GB. Second, my testing occurred before Raidtec's addition of Win2K support, and Raidtec sent an early build of software that supported the joining of Win2K domains. This build let me join a Win2K domain, but the RAIDserver V claimed to be a Domain Master Browser, causing a conflict with the true Domain Master Browser (i.e., the PDC-emulator DC) when the RAIDserver V was online. Raidtec reports that production builds of the software don't have such problems. Raidtec will likely have full Win2K support working by press time.
Database on NAS
Microsoft officially supports only Network Appliance's NetApp filers to host Microsoft SQL Server databases. Microsoft doesn't recommend NAS units for Microsoft Exchange Server database storage.
The reason for this limited support is simple. All access to data on NAS devices relies on the robustness of the TCP/IP protocol suite. The TCP/IP protocol tries to guarantee delivery of data across a network to the intended application but isn't always successful. Applications such as SQL Server and Exchange Server rely heavily on the complete integrity of their databases, and the current state of TCP/IP over Ethernet often falls short of providing complete data integrity.
Three Tiers
These five systems fall naturally into three categories. In the low-end category, both the NetFORCE 1500 and the RAIDserver V offer easy-to-use Web-based administration tools, although I'd give the ease-of-use edge to the NetFORCE 1500. The RAIDserver V is more expandable, but I experienced problems with Win2K domain participation, and the unit's 2GB file-size limit tripped up my testing.
In the high-end category, the NetApp F840 and the MetaStor 4300 are UNIX-based systems with multiterabyte storage capabilities. Both systems were more difficult to administer than the lower-end units, but the higher learning curve is to be expected, given the products' expansion capabilities. From a performance perspective, the NetApp F840 was the star of this show, providing performance that was double the speed of my 4-way reference server.
Finally, I'm placing the TaskSmart N2400 in a class by itself. Because the system is based on Win2K AS, Win2K administrators will find that it's the easiest of the systems to learn and manage.