By Jack Fegreus
With the price per megabyte for high-capacity Ultra ATA storage devices slipping below a penny per megabyte, how is it that Maxtor has found a way to sell storage at a 250% markup and make that seem like a bargain? The answer, of course, is that Maxtor has moved into the burgeoning value-added space of network-attached storage (NAS).
Even HP, which no longer manufactures drives, has moved solidly into the NAS space. While Quantum and Maxtor have initially targeted the small/home office (SOHO) business space, HP is aimed squarely at the IT market for high-end departmental storage servers.
As we noted in February's review of Hewlett-Packard's SureStore HD Server 4000, the goal for all NAS vendors is to provide a storage appliance that can be unpacked, plugged in, and used as easily as a toaster. HP exclusively targeted corporate IT with departmental-scale storage servers with features such as redundant power supplies and hot-pluggable SCSI-2 drives in a RAID-5 configuration.
We measured throughput for sequential reads at the client workstation. For applications, the critical read size is 8KB. At this size data transfer, streaming throughput from the MaxAttach was on the order of 5.7MBps.
In distinct contrast, Maxtor has introduced its first- and second-generation products at the SOHO space. Still, just like HP, Maxtor has exclusively targeted the Windows 9x and NT/2000 networking markets with all communications done via TCP/IP.
Like any good appliance, the new MaxAttach 2.0 has a very tiny footprint. In fact, this NAS device is literally not much bigger than an externally powered disk drive. There are three basic configurations for the MaxAttach, based on storage capacity: 20GB, 40GB, and 80GB. The devices are priced at $995, $1,595, and $1,995, respectively. For the Data Storage Magazine tests, we benchmarked an 80GB MaxAttach 2.0.
The MaxAttach 2.0 is powered by a classic P5 Intel Pentium processor, which is running Open-Source Free BSD Unix. Nonetheless, this is only the starting point for Maxtor. They have modified the native Free BSD file system to allow this NAS appliance unit to improve throughput performance with caching and to improve reliability by providing self-repair and data recovery capabilities.
Naturally, all of the OS pyrotechnics are totally masked from the end user. All management of the MaxAttach is via a ubiquitous web-browser interface. The website on the MaxAttach is very simple and with three or four clicks, the NAS server can be fully configured and put into service.
Actually, the real importance of the intranet-based management interface is as an indicator of future products. For example, it would require little or no effort on the part of Maxtor to turn the MaxAttach into a web-server appliance.
The SOHO-oriented MaxAttach devices use very high capacity Ultra ATA drives. These drives can be configured as two separate JBOD devices or as a single logical drive by either spanning the two drives or mirroring the drives in a RAID-1 configuration. In the latter configuration, both user and server data is mirrored so that in the event of a disk failure, the MaxAttach unit can boot and run from the remaining drive.
The unit tested by Data Storage Magazine had a capacity of 80GB: two DiamondMax 40 series drives. At a time when most drives being sold into PCs have just one platter, the DiamondMax 40 series drives have four platters and eight heads.
Spinning at 5,400rpm, these drives sport a sub-9.0 ms average seek time and augment their performance with a fast 2MB 100MHz SDRAM cache buffer and an UltraDMA 66 interface. They also employ Maxtor's DualWave multi-processor controller, which the company touts as providing a 10x boost in processing host commands.
Testing was done in a Windows 2000 domain environment. Our client was a 200MHz P6-based Dell OptiPlex GXpro workstation running Windows 2000 Professional. We used the Nova Technica Disk I/O benchmark (available for free download at www.novatechnica.com) to measure streaming read/write throughput and File Load benchmark to simulate query access in a database environment.
Setting up in a Windows 2000 environment, we ran into two problems with the MaxAttach, which will be addressed in an upcoming patch by Maxtor. We could not change folder names from a Windows 2000 client and, more importantly, we could not log into the administrator web site to configure the MaxAttach device. A Java login applet could not register our password from Windows 2000.
A Windows 98 client demonstrated none of these problems. From that client, we configured the MaxAttach server as a RAID-1 device with 40GB of mirrored storage available.
In a single-user scenario, streaming throughput from the MaxAttach was stellar. For all read sizes-4KB through to 64KB-throughput from the MaxAttach exceeded that of the SureStore HD Server by roughly 55%. While this performance is quite impressive, the ability of a storage server to scale as the number of simultaneous users increase is often even more important at many sites. As the number of users trying to access data increases-even in a simple file-sharing scenario-their access patterns tend to follow a distinct distribution pattern. To model this scenario, we used the database simulation option that is part of the Nova Technica File Load benchmark.
In this benchmark, 50% of the I/Os are randomly dispersed over 25% of a large test file to simulate index access. The remaining 50% of the I/O load is then randomly distributed over the remaining 75% of that simulated database file. As database daemon processes are added, they compete intensely for access to the index area. As the number of daemons increase, a highly effective I/O subsystem caching scheme as well as a file system striped over a maximum number of spindles is essential for taking the pressure off the drive's command queue and extending the responsiveness of the I/O subsystem.
With just simple mirroring and no striping, we did not expect the level of performance demonstrated when streaming data to a single user. Nonetheless, when we scaled the client load, the MaxAttach not only held its own against the SureStore HD Server, but it outperformed the larger server starting with 10 clients and continuing up through 50 simultaneous database clients-each of which issued its own stream of several thousand I/O requests-without any difficulty. It would appear that the caching module Maxtor added to the Free BSD distribution along with the DualWave multi-processor controllers on the drives paid off in very measurable performance.
Finally, we also looked at Reflect-It, another added-value software option that Maxtor sells for $79. There is a Reflect-It server resident on the MaxAttach, and this client software permits a desktop system to back up files on any and all of the desktop system drives to the MaxAttach unit. The software, which worked perfectly from Windows 2000 clients allows for on-demand and scheduled full and incremental backups of data to the NAS server.
There is also a mode called real-time mirroring, which is designed to make simultaneous updates to the MaxAttach as files are changed on the desktop. Unfortunately, this is implemented as an application and not a pseudo driver.
As a result, when a file changes, all files being monitored are checked, representing an incredible waste in overhead. Scheduling a daily check in the wee hours of the morning makes infinitely better sense.