Multimedia
Blogs
Forum
Affordable HD
Whitepapers
Newsletters
Events
Advertisers
Blogcast
MillimeterBreaking the Speed Limits
Jul 1, 1997 12:00 PM, by Steve Mullen
What Drives Your Disk Drive? A Guide to the Latest Developments in Digital Storage Technology; Also, Tips for Getting Better Performance from Your Storage Devices
While it's conceivable that one could be too rich and too thin, it's much harder to conceive that one could have too much fast mass storage. Disk manufacturers and their OEMs have developed a new set of advanced technology products that should meet the increasingly complex requirements of power users in the production and postproduction industry.
This story will examine the storage technology story behind these new products, because each technology advance represents an option one may want to adopt. The primary focus will be high-capacity hard disk systems that connect directly to a nonlinear editing system. However, networked disk systems and disk-based VCRs will also be examined.
AV Hard Disk Design Platter Rotation Speed With one exception, AV disks (AV Drives are optimized and calibrated for video playback without dropping frames) have a rotational speed of 7,200 rpm. At this speed, the outermost tracks can transfer data to and from the disk's cache at nine MB/S. (The outermost partition is the one you create first. It appears at the top of a partition list.) The innermost partition can transfer data to and from the disk's cache at six MB/S. From a conservative viewpoint, the capture and playback speed of a 7,200-rpm drive is therefore limited to six MB/S. (In this story, "MB/S" or "MB/sec." refers to megabytes-per-second, unless otherwise indicated.)
There are two ways to increase the potential transfer rate. First, either more disk platters can be used, so that only the outermost half of each platter is used (a very expensive solution), or track density can be increased, so that only the outer portion of the disk surface is utilized for storage (a solution that requires advanced read/write and tracking technology).
Second, the rotational speed can be increased. Seagate has pursued the second alternative with its family of Cheetah drives. These are the world's first 10,000-rpm drives. For Seagate, which introduced the world's first 7,200-rpm drive-the Barracuda-the Cheetah is a natural extension of their technology. The faster rotational enables data rates of 11.3 MB/S to 16.8 MB/S. In many cases this level of performance eliminates the need for creating a disk array. The Cheetah is a 3 1/2-inch form factor drive and is available in 4.55GB (one inch high) and 9.1GB (1.6 inches high) capacities.
Seek Times Applications that play back QuickTime movies assume that the movie will be stored on disk as a series of sequential data blocks. If these conditions are met, playback performance is primarily a function of rotation speed and track-to-track seek times. The Cheetah's high-speed and less than one-millisecond seek times should provide very high QuickTime movie playback performance.
Advanced NLE software, however, does not require a movie to be made, thereby avoiding both wasted time and disk space. These applications require perfect playback from a series of files scattered all over the disk. Obviously, high rotational speed allows very high data bursts from each file to the application. And high rotational speed also lowers "average disk latency." The Cheetah's average latency is under three milliseconds. This minimal rotational delay, plus a quick response from the Cheetah's advanced servo technology, reduces average read seek time to under 7.5 milliseconds. Seagate drives are rated to have a 1,000,000-hour MTBF ("mean time between failures," a manufacturer's estimate of disk life before a head crash) and a five-year warranty.
Head and Cylinder Structure The key to all the new advanced AV drives from Seagate and Micropolis is Magneto-Resistive (MR) heads that offer better off-track capability and an improved signal-to-noise ratio, thus providing greater tracks per inch. This head design, plus Partial-Response, Maximum Likelihood (PRML) technology incorporates a series of read-channel algorithms that are far more accurate than traditional peak-detect read technologies.
Seagate drives feature embedded servo technology that eliminates the need for thermal recalibration (T-CAL). Micropolis drives use a second generation hybrid servo system that offers the robustness typically available only with drives that have fully embedded servos, plus the high performance of dedicated servo drives. The hybrid servo also eliminates the need for T-CAL.
The Micropolis Tomahawk features a unique Expedited Error Recovery (EER) technology specifically designed for digital audio/video applications. To eliminate data gaps that result in dropped frames or sound pops, the EER system incorporates a sophisticated hardware and software on-the-fly ECC capability that improves the drive's margin for error correction by up to four times over competing drives.
The Tomahawk, which has a 3.5-inch form factor, is available in 4.55GB (one inch high) and 9.10GB (1.6 inches high). It features a 7.9 millisecond average seek time. Micropolis drives are rated to have a 1,000,000-hour MTBF and a five-year warranty.
Because Seagate's 10,000-rpm drives were not available at the time this story was being written, second-generation Barracuda drives were considered instead. These drives use the same technology as the Cheetah and are competitive with the new Micropolis Tomahawk AV Gold drives.
Cache Size The Tomahawk drives feature a two-megabyte, multi-segmented cache that enables Micropolis to guarantee an uninterrupted sequential data transfer rate of 6.5 MB/sec. This is a 63 percent increase over previous AV drives. Seagate drives offer a 512Kbyte multi-segmented cache with drives that have an UltraSCSI interface. A 2,048Kbyte cache is optional. Drives with a Fiber Channel port, include a 1,024Kbyte cache.
Disk Connection Path SCSI and EIDE are the most common technologies used to connect disk drives to a computer's PCI bus or motherboard. SCSI is the faster of the two connection schemes. The Mac has always used SCSI, while it is an option on high-performance PCs. The original SCSI connection, called SCSI-1, offered a transfer rate of up to five MB/sec. Even today's Power Macs use a SCSI-1 for the external SCSI port.
Fast/Narrow SCSI-2 To increase performance, SCSI technology was enhanced, thereby creating SCSI-2. SCSI-2, usually called Fast SCSI, offers twice the potential data transfer rate. Fast mode simply cuts the synchronous transfer timings in half. The most common form of SCSI-2, called Fast/Narrow, has a maximum transfer rate of 10 MB/sec.
The Narrow designation indicates that data is passed over an eight-bit path-the same as SCSI-1. Power Macs have a Fast SCSI-2 port internally. It's used to connect the Mac's hard disk to the motherboard bus. Unfortunately, Apple has never provided an external connector for this bus. If your PowerMac can handle the heat generated by a 7,200-rpm drive, it can be mounted within your Mac and connected to this bus. Usually, however, PCI SCSI-2 adapters have been used to support high-speed external hard disks with both Macs and PCs.
Fast/Wide SCSI-2 By expanding the eight-bit data path to 16 bits, the potential transfer rate is increased to 20 MB/sec. Most AV drives are available with a Fast/Wide connection at a very small price increment. Despite the greater maximum performance offered by a Fast/Wide bus-sustained disk transfer rates are not doubled. Rather, the greater bandwidth easily accommodates the 12 MB/sec. or greater burst transfers generated by 7,200-rpm drives. For this reason, Fast/Wide drives offer AV applications consistently high sustained data transfer rate.
Ultra/Narrow SCSI Simply put, UltraSCSI is a third revision SCSI standard. Ultra/Narrow SCSI offers the same 20 MB/sec. limit as does Fast/Wide SCSI-2. Because Ultra/Narrow offers no dramatic improvement in disk transfer capability over Fast/Wide-it's not particularly valuable.
Ultra/Wide SCSI Ultra/Wide SCSI, with its 16-bit data path, offers up to a 40 MB/sec. maximum transfer rate. I tested an ATTO UltraSCSI ExpressPCI single-channel adapter and a pair of Seagate ST34371W 4GB UltraWide drives, all supplied by MicroNet as a DataDock 525 subsystem, and measured outer-partition average read and write performance at 16.5MB/S.
These results were obtained with data transfers from 500Kbytes to 1000Kbytes. Performance, from the inner-partition, was 13.5 MB/sec., with 14.5 MB/sec. average read and write performance, from outer-partition.
See the "Performance Testing" sidebar (page XX) for information on testing your disk's performance.
Differential SCSI The majority of hard disks sold have "single-ended" signal connections. This type of connection sends signals via a single line. The signal return path is via a common ground lines that are connected to the chassis of each device. If the devices are far apart there is a good chance the chassis ground potentials will be different. This difference, which could be constant or varying, can interfere with the data signals.
For this reason, single-ended SCSI cables are limited to six meters. Differential connections use a pair of lines to carry each signal. Because no common grounds are involved, differential connections are far less sensitive to differences in chassis ground potentials. For this reason, cable lengths up to 25 meters are possible.
SCSI Accelerators There are half a dozen SCSI accelerator cards on the market. Naturally each one boasts a set of features-RISC processors, on-board RAM, DMA operation, etc. Two of the most popular accelerators are marketed by Adaptec (2940UW) and ATTO (ExpressPCI).
SCSI accelerators are available in single- and dual-channel versions. Dual-channel boards offer two SCSI channels. The best performance is achieved when multiples of four drives are used with dual SCSI channels. Dual-channel adapters are of particular value for users who require over a 20 MB/sec. data stream to feed dual codec boards that support real-time effects.
Array and Non-Array Solution Any even multiple of disks can be configured as a "disk array." A disk array uses a more advanced form of a SCSI software (or hardware) that supports the simultaneous operation of pairs or quadruples of disks. RAID (Redundant Array of Independent Disks) software enables an application to treat each pair (or quadruple) of disks as a single disk.
The magic of RAID is that the performance of a pair of disks is almost doubled over the performance of a single disk. When multiples of four drives are used, performance can be almost quadrupled. RAID software is available from Adaptec (REMUS), ATTO (ExpressRAID), FWB (RAID Toolkit), MicroNet (DiskWorks), and MegaDrive (MegaDrive RAID).
The new UltraSCSI Micropolis and Seagate drives are so fast that you may not need to implement an array. These drives are of particular value to PC users because Windows '95 lacks software to create arrays. (Windows NT includes RAID SCSI drivers.)
I tested a single Micropolis 3391W 9GB drive, mounted in a DataDock 525 chassis, connected to an ATTO ExpressPCI. Average read and write performance both reached 8.4 MB/sec. from the outer partition. Average access time was 10 milliseconds.
Remarkably, read and write performance from the inner-partition was equally good at 7.9 MB/sec. These results were obtained with data transfers from 500Kbytes to 1000Kbytes. For individual 4GB Seagate drives, the average read and write performance from the outer partition was measured at 8.3 MB/sec. and 9.1 MB/sec., respectively.
Fibre Channel Fibre Channel is the hottest alternate to UltraSCSI-although SSA also has strong support. Fibre Channel (FC) uses a laser driven fiber optic cable to move data at up to 100 MB/sec. Disk drives with an FC interface are connected in an Arbitrated Loop. Up to 10 km of cable can exist between each FC-AL device with no loss in data throughput or reliability. Up to 126 devices can be connected in a loop.
Unfortunately, FC-AL can't be added to your current SCSI disk drives. Seagate FC-AL drives include Exclusive OR (XOR) technology that enables drives to directly communicate with each other to confirm data integrity when the array is configured to provide "fault tolerant" operation. Traditionally the RAID controller (whether software or hardware) had to perform multiple time consuming operations that included reading data written to drives, performing an integrity check on this data, generating new redundancy codes, storing the codes on the drives, and then writing new information to the disks for each block of data.
Seagate's XOR functionality enables the drives to perform these actions independently and automatically. Box Hill Systems is now shipping 9GB XOR drives for Windows NT.
Choosing a Drive Seagate and Micropolis drives are purchased and resold by a number of other companies. Traditionally, resellers have hidden the source of their drives. Now, however, most resellers clearly specify in their product literature both the manufacturer and model number of the disk drives they use.
In order to provide product differentiation, resellers must focus on issues other than the drives themselves. Differentiating features fall into seven categories: hardware (e.g., power supply, cooling, termination), software (e.g., formatters and disk drivers), convenience (e.g., drive ID selection), documentation, warranty, customer support, and price. Resellers include: MegaDrive (Enterprise E-2 and E-8), MicroNet (DataDock 525 and 7000), FWB, and Optima-as well as a host of mail-order companies such as APS, JEMS, and ProMax.
Video Networks In large postproduction facilities, there are likely to be multiple NLE systems. There are also likely to be graphics and animation workstations. Except where these systems are rented to individuals, there is the necessity of sharing sound and the images between all these systems.
Sharing video files is complicated by the fact these systems are very likely to be supplied by multiple manufactures and have different operating systems. The traditional solution is hot-swappable disk drives. When data is needed by different people (or in separate rooms), the necessary disk drives are physically moved from one system to another. Obviously, this provides a very limited resolution to a demanding problem. There are two solutions that are far more sophisticated.
The first enables the disks attached to each system to be shared with the other users. Transoft is a key supplier of software and hardware that implements a distributed network solution. Their StudioBOSS solution enables the users to simultaneously access large data files at 200 MB/S (duplex) on PCI-based Macintosh, Silicon Graphics, and Windows NT computers.
The Transoft network uses Fibre Channel Arbitrated Loop (FC-AL) technology. Transoft also supplies 72GB and 180GB storage systems that use Seagate drives.
The second solution uses a client-server model where most of the facilities storage is contained in a server. MountainGate's CentraVision system is an example of this type of solution. The storage unit contains either four (36GB) or eight (72GB) disk drives organized as a four-way striped array. A FC-AL connection is made to a FC Hub that allows eight computers to be connected. (Alternatively, Hubs can be cascaded to support up to 126 computers.) The MountainGate solution uses the StudioBOSS software.
Disk-based VCRs: DDRs Sierra Design Labs pioneered the development of disk-based VCRs. These devices appear to other equipment to be a digital VCR. They are controlled via RS-422 ports and respond to the usual SDL/Sony/SMPTE VCR control commands. Source timecode is available and the decks can be used for frame accurate editing. Video is input and output as either or both parallel and serial digital data. Audio data is input and output as AES/EBU digital data.
An analog reference input is provided as well as an analog output that displays the video image overlayed with menu information. The video information is stored on a disk array as uncompressed 4:2:2 component data using either eight-bit or 10-bit samples. Both 525-line (SMPTE 125 and SMPTE 259M) and 625-line (CCIR 601 and CCIR 656) formats are supported. Sierra design's Quickframe 96 can hold up to 95 minutes of video with eight-bit sampling and 76 minutes with 10-bit sampling.
The Quickframe series also can operate as a server using a variety of connection options. Ultra-Wide SCSI ports allow direct connection to a computer system. Using Sierra's NFS Server, computers can be connected to servers via 10Base-T and 100Base-T Ethernet and via ATM. Sierra plans to add FC-AL support in 1997.
Many other manufactures have brought DDRs to market. These include QuVis and Pluto Technologies, HP, Tektronix, Ciprico, Mercury, Fast Forward, and Avid. Clearly DDRs-that also function as servers-are becoming the hot new product for both individual and shared-user postproduction applications.
AV disk drives have a number of operational parameters that can be set to define how they will cooperate with the host computer to transfer data. When these parameters are correctly set, transfer data rates will be optimal. Many mail-order drive venders ship drives to you with the default values established by the OEM manufactures of the drives.
These settings may, or may not, be optimal. The problem is to either obtain, or through experimentation discover, the correct values. To set parameters you will need a program like FWB's World Control, MicroNet's DiskWorks, or Disk Control supplied with REMUS. Begin first by adjusting those values that determine disk write performance. Adjust for the maximum capture rate without causing an error to be reported. Be sure to fill the entire innermost disk partition because this partition always offers the lowest capture performance.
Next, playback this portion. If you get dropped frames, adjust those parameters that effect disk read performance. If your best tuning still yields dropped frames, you have no alternative than to lower the capture data rate. This time consuming procedure must be repeated until the entire inner disk partition can be played back without dropped frames.
If, when you perform your tests, you find the resulting capture rate too low, test the next partition out from the innermost partition. Then reserve the innermost partition for storing static graphics files. Do not store audio files that will be played back with video in this partition-although other audio files can be stored here.
If you are working with products such as the Media 100qx, DPS Spark, or FAST's DV Master, where you must make a QuickTime or AVI movie, there is a way to increase movie quality without introducing dropped frames. First, reserve the outermost disk partition for the QuickTime (2GB) or AVI (1GB) movie. By doing so you will place your movie in the fastest portion of the disk.
Now, follow the same testing procedure described above-but test this partition. When you have determined the maximum capture rate than can be played back without dropped frames from the "movie" partition, check the innermost partition to be certain it can support this capture rate. If not, you can either lower the capture rate or not use the innermost partition for capture. (See the "Performance Tuning" sidebar.)
No matter what testing technique you follow, the following mode page parameter specifications are a reasonable place to start experimenting.
*AWRE-Automatic Write Reallocation Enable. If a write error is detected this option, which takes time to execute, will reallocate any bad blocks encountered. This can be set "off" because errors with digital video and audio data are treated as no more serious than "dropouts" on tape.
*ARRE-Automatic Read Reallocation Enable. If a read error is detected, this option, which takes time to execute, will reallocate any bad blocks it encountered. This should be set "off".
*EER-Enable Early Recovery. If your drive supports this option it should be turned "on". This will apply the error code correction as soon as possible.
*RC-Read Continuous. If your drive has on-the-fly hardware error correction, you should turn it "on". This will override and disable AWRE, ARRE, EER, and Read and Write Retries. Hardware error correction doesn't waste any time.
*Read Retry-If an error is detected during a read the operation is retried causing, with a 7,200-rpm drive, an eight-millisecond rotational delay. To eliminate this potential delay, set the parameter to "0." If you want to give the drive a chance to recover from an error, set the parameter to "1" or "2." Set the parameter to "15" if you want greater safety. Do not store valuable data on the disk if you set the value under "15."
*Write Retry-If an error is detected during a write, the operation is retried, causing an eight-millisecond delay. To eliminate any delay, set to "0." If you want to give the drive a chance to recover from an error, set the parameter to "1" or "2." Set the parameter to "15" if you want greater safety. Do not store valuable data on the disk if you set the value under "15."
*Disconnect/Reconnect Parameters Buffer Full Ratio-This value defines, in data blocks, how full the cache must be prior to reconnecting to the host for another read operation. Set to "32" of 256 blocks thereby setting the ratio to 0.125.
*Buffer Empty Ratio-This value defines, in data blocks, how empty the cache must be prior to reconnecting to the host for another write operation. Set to "224" of 256 blocks thereby setting the ratio to 0.875.
*WCE-Write Cache Enable. When enabled, the controller will return a command "done" message as soon as data is written into the cache by the host. Enabling the cache will increase your sustained writes rate substantially.
*RCD-Read Cache Disabled. When the read cache is enabled, the controller will return a command "done" message as soon as data is read from the cache. Do not disable the cache.
*Minimum Prefetch-This value specifies the number of 512 byte blocks to prefetch into the cache from the disk. Set the parameter to "0." This doesn't turn off prefetch. It simply allows prefetching to be interrupted when a read request is received from the host.
*Maximum Prefetch-This value specifies the maximum number of 512 byte blocks to prefetch into the cache from the disk. Set the value to "255" unless your drive will allow no other value than "0."
*Number of Cache Segments-Set the value to "3" unless your drive will allow no other value than "0."
MicroNet sent me a copy of their DiskWorks software which they claim will take the time-consuming testing out of tuning disks for maximum performance. First I tested the DataDock 8GB Seagate array after using DiskWorks Auto Prepare function-choosing "Video Production" as the target application.
Typical read and write performance for the array, as measured by the DiskWorks benchmark, was 16 MB/sec. and 9 MB/sec., respectively. Then, using my Media 100 at a video quality setting of 150kB/F, I captured and played back four minutes of video. Next I activated the Drive Tuner function of DiskWorks. By clicking on the "extract data from log" button, the typical data transfer size from my Media 100 session-768Kbytes-was obtained.
Next, I let Drive Tuner test the array to determine the optimal Mode Page parameters for my application. It took four hours to perform the tuning operation after which the DiskWork's benchmark was run again. Typical read performance now measured 16 MB/sec. while typical write performance measured 19 MB/sec.
These tests indicate that automated tuning has the potential to perform a valuable function. Just for comparison, I set the Mode Page parameters to my "best estimate" values as specified above. The DiskWorks benchmark function measured typical read performance at 16 MB/sec. and write performance at 17 MB/sec. The automated process yielded a difference in one parameter-the Buffer Empty Ratio was set to only 25.
Thus, I chose to use the parameters established by DiskWorks. For the Micropolis 9GB, I also chose the Drive Tuner settings over my recommendations because they raised write performance to 9MB/S. Once again the automated process yielded a difference in only one parameter-the Buffer Empty Ratio was set to 130.
Continue the discussion on “Crosstalk” the Millimeter Forum.
