Post a CommentFeature
NDAS undressed: dissecting a NAS substitute
Can a thermal-management problem explain the early death of this storage device?
By Brian Dipert, Senior Technical Editor -- EDN, 5/11/2006
| Photo gallery Click on any of the images below to see an extreme closeup. You'll also find links to the images within the text of the article.  First-generation NetDisk  Second-generation NetDisk  First-generation NetDisk circuit board  Second-generation NetDisk circuit board  Underside of first-generation NetDisk circuit board  Underside of second-generation NetDisk circuit board  Inside of top lid of first-generation NetDisk enclosure  Inside of top lid of second-generation NetDisk enclosure  Third-generation NetDisk circuit board |
Ximeta's NDAS (Network Direct Attached Storage) NetDisks undercut "pure" NAS (network-attached-storage) drives in price. But in order to do so, they shift some of the storage-processing burden to the devices that connect to them. You need to install specialized software on each LAN client that you want to be able to read or write the drive(s). (Incidentally, this software has historically been flaky in my experience, especially when I attempted to access the NetDisks over a wireless connection.)
Where did the NetDisk problem lie?
- The client-side software?
- The network link between my PC and the NetDisk?
- The Ethernet-to-Parallel ATA connection within the NetDisk?
- Or the NetDisk's HDD?
The first scenario was easy to rule out; none of the computers on my LAN could connect to the faulty NetDisk. To test the second and third possibilities, I disconnected the NetDisk from the LAN and direct-attached it to my PC over USB. Still nothing. I was left with the highly likely scenario that the HDD itself was faulty, although the circuitry inside the NetDisk that blends the USB and Ethernet system connections into a common PATA link to the HDD was still suspect.
|
A true NAS unit incorporates a microprocessor (standalone or as a core within a system-on-chip), running an operating system such as Linux, in conjunction with many dozens or hundreds of megabytes of nonvolatile and volatile system memory. NDAS makes do with a proprietary 80-lead TQFP ASIC (whose myriad functions you can decipher here), along with two serial EEPROMs, 2 kbits (Microchip Technology 24LC02B) and 16 kbits (Atmel 24C16) in size. The latter is socketed and, in the first-generation unit, sports a hand-written ROM code sticker! The 2-kbit device is socketed in the first-generation unit but soldered down in the second-generation product.
A Cypress Semiconductor CY7C68300 USB 2.0-to-ATA/ATAPI bridge chip and an Intel LXT972A 3.3V 10/100 Ethernet transceiver IC tackle the system-interface duties, the latter in conjunction with a HALO TG110-5050N2 magnetic isolation module and Pilkor PCX2 337 275V-tolerant, electromagnetic interference-suppressing capacitor. Two Texas Instruments HCT245 octal bus transceivers link the HDD to the remainder of the NDAS. Additional analog and passive circuitry includes a 1A low-dropout voltage regulator; 24-, 25-, and 50-MHz oscillators; and a protection diode. Illuminated LEDs indicate active power and HDD access.
Back-panel connections support USB, wired Ethernet, and dc power. As you can see in this close up of the circuit board, first generation units relied on user-controlled manual switches to toggle between USB and Ethernet. Second-generation devices automatically select the proper system connection. Also missing from the second-generation device is a board-mounted jumper. A look at the undersides of the first- and second-generation boards reveals no active circuitry, only passive components, solder points for through-hole ICs, and traces.
When I pulled the drive from the enclosure and attempted to mount it on my Windows PC via a USB2-to-PATA adapter, it again failed, identifying the HDD as the ultimate culprit. But what caused the HDD to fail? Before I reveal my theory, let me first give you a few important tidbits of information concerning my setup. First, I acquired the older, first-generation NetDisk brand new, whereas I picked up a factory-refurbished unit of the newer, second-generation unit.
The first-generation NetDisk contained a Maxtor 160-Gbyte HDD, whereas the second-generation unit contained a Samsung HDD of the same capacity. I can't tell from looking at the Samsung HDD whether or not it, befitting the NetDisk's refurbished status, was new or had been used prior to coming into my possession. Both HDDs are 7200-RPM models with 2-Mbyte buffers—a baffling feature-set combination. I presume that by going with a 2- versus 8-Mbyte cache Ximeta cut costs, especially considering that NetDisks don't support Gigabit-Ethernet connectivity. But why, then, did Ximeta spend the extra money for 7200-RPM HDDs versus 5400-RPM variants?
Ximeta's 7200-RPM decision has potential cost impacts. It also likely had thermal impacts, which brings us to the crux of my cause-of-failure theory. Ximeta's documentation encourages stacking of multiple NetDisks. So do the rubber feet on the bottom of each enclosure. In my case, the older first-generation drive was on top, with the newer second-generation drive below it. Each drive ran quite hot, which is not altogether surprising given that there's no fan inside the enclosure, or even any vents that, if they existed, would allow air to passively enter and exit the case.
Look again at the overview shots of the first- and second-generation enclosures. Notice the discoloration of the internal plastic in both cases? Now look at the insides of the first- and second-generation enclosures' top lids, and notably at the additional discoloration in the second-generation case. I can't say for sure that this discoloration wasn't present from the beginning of the enclosures' lives, because I didn't crack the enclosures open until I had a failure. But I suspect the discoloration was heat-created, and that because the lower NetDisk in the stack wasn't exposed to ambient airflow (since it had an equally hot drive sitting on top of it), the intensity of the captive heat exposure accelerated its time-to-failure.
Last-minute addition: In the process of answering my questions regarding various NetDisk constituent components, Ximeta realized that both of my existing NetDisks were out-of-date, hardware-wise, compared with currently shipping units. Here's a closeup shot of the latest-generation circuit board. Several items of particular note come to my mind:
- The altered parts list and reduced overall parts count
- The uptick of the Ximeta ASIC version number, from v1.0 in first- and second-generation NetDisks to v1.1 in this latest-generation iteration, and
- The uptick of the NetDisk circuit-board numbers, from v1.1 in the first-generation unit, to v1.4 in the second-generation unit, and finally to v1.8 in this latest-generation unit.
Acknowledgement
Thanks to Gartner Dataquest analyst Jeremey Donovan for providing me a copy of his late November 2004 NetDisk teardown report, which helped me identify several of the systems' components.
Author information
You can reach Senior Technical Editor Brian Dipert at 1-916-760-0159, bdipert@edn.com, and www.bdipert.com.
Editor's note
The above is an extended version of an article that appeared in shorter form in the print edition of EDN. This PDF file shows the printed version.
