Explode the Enterprise Info Glut with TNT
Meridian`s NT-based CD-ROM/DVD device serves discs to the masses.
By Barrie Sosinsky
Today some cool stuff comes on either CD-ROM or on DVD, much of it essential tools for the enterprise. All manner of reference material is available, from the technical manuals for the biggest jumbo jets to databases of information for any specific industry. Since you can also burn your own CD-ROM discs rather inexpensively these days, your own company`s data can be in this format as well.
There`s a great need for companies to make information on read-only media available across a network. Since the material is often expensive, this needs to be done in a way that shares the resource and keeps the equipment necessary to access the material reasonable. For these reasons, CD-ROM/ DVD-ROM servers have become popular in organizations that manage large volumes of information or shared resources.
IT\IS Labs looked at an NT-based CD-ROM/DVD server from Meridian Data. This system is a hardware/software package that plugs directly into your network as an NT Server system. Meridian also offers servers using Novell NetWare and other operating systems, but this unit is the one a BackOffice shop would be most interested in.
From a hardware point of view, setting up the TNT Server is straightforward. It comes with NT Server 4 SP3, and once you define your networking protocols and enter the server`s network address, the server is on-line. Since the server manages multiple hard drives and uses them to cache CD-ROMs or DVD discs, you must decide how to administer those drives.
In the unit we reviewed there were four 6GB drives, with Disk 0 storing the operating system in a 500MB partition, and all four drives participated in a striped set with parity. Meridian set up the server`s disk this way under the assumption that you would want RAID`s fault tolerance to recover data should a disk fail. If you are using cached data that isn`t simply read-only, you may want this configuration even though performance suffers.
The CD Net software package that manages your media requires both server-side and client software installation. This part of the installation is easy. CD Net installation can install three small utilities: CD Manager, CD User Administration, and CD Net Tools. You can also install an application called Slingshot! that sets up your management interface over an intranet. TNT Server`s CD Net Server software will also manage a jukebox, although we did not test that feature.
During setup, you are shown what disks are on your system, and can mark those drives as being available for caching. You can also specify how many CD-ROMs you wish to cache, and how much of the disk space you can consume on a drive before the drive is considered full. Most administrators leave from 10% to 15% of the disk empty.
Each CD-ROM you cache will consume between 600MB and 650MB of disk space, and a DVD-ROM disc can consume about 8GB of storage. That being the case, you don`t typically get many DVDs cached on a system. With the 9GB hard drives now available, you can get one or two DVDs cached. On the four 6GB disks on our system, we had a theoretical capacity of 23 CD-ROMs cached to disk, in addition to the 7 CD-ROMs and 1 DVD disc that could be stored in the physical drives themselves.
One of the problems you face installing and running a library of read-only media is that each disc or cached disc consumes a drive letter. Since NT allows only 26 drive letters, you would be limited in your system`s capacity. To circumvent this problem, CD Net creates a virtual file system and can trim the drive letters of any additional disks you add beyond a drive letter you specify, typically configured as the G or H drive. As you add additional disks to your system (either physical or cached), those disks appear as folders in the virtual file system.
If you`ve set up your system to auto- recognize physical disks, then the disks become folders that can be network shares. Users can open that network share and have access to the data in it. If that disk has an application associated with it, then a network user can install the application locally on the client and run from the data contained on the disc on the TNT Server. Running off a share will operate whenever the application respects UNC conventions. For example, Encarta 98 on DVD worked correctly; although we were told that earlier versions did not.
Whenever you work with discs that aren`t UNC compliant, you must set up a volume in CD Manager and have a shortcut properly configured in a database you create in CD User. In CD Net, a volume is a physical CD in a drive, a cached CD, or a CD imported to the jukebox shelf.
You can also create a volume group that spans several volumes. Putting a related set of CD-ROMs into a group volume allows the application using those discs to automatically find any disk in the disk set without prompting the user. NT users and groups can be assigned access to a volume or volume group, a very important feature on a network for security.
In configuring the TNT Server, most of the time is spent using CD Manager and the CD User applications. In CD Manager, you set up volume labels, cache read-only media to hard drives, and perform setup tasks. Volumes can be set up so they can be accessed by only a certain number of users, to comply with a product`s licensing requirement. CD User lets you create a database of applications with the correct pointers that can be accessed by network users.
Working with CD Manager and CD User was somewhat confusing from an administrator`s point of view. Central features, such as caching a drive, aren`t well described in the documentation or the on-line help system. The documentation requires you to read carefully to understand something that should be part of a simple step-by-step task. However, when we called Meridian`s technical support line, our calls were returned promptly and the technical support was good.
The range of features TNT Server`s software supports will impress administrators, once they know the product. The tools provided do the job. CD Net even comes with a nifty utility called CD Net Tools that provides you with an analysis of the TNT Server`s performance, usage, and current activity.
When you only have a few users accessing a CD-ROM or DVD drive in a drive bay, then you can run those discs as physical disks. The real benefit that the system provides is the ability to cache volumes. Caching a typical CD-ROM takes about 20 to 30 minutes, and the system allows you to cache two or more discs at once--a real time saver when you are loading an entire new disk set.
Sequentially Seeking Speed
To test performance, we ran a drag-and-drop file copy of a folder containing 27.9MB of data across the IT\IS Labs 100MB Ethernet network. Performance locally on a 12X CD-ROM on a client machine took nearly twice as long as running the same program from the drives on the TNT server. Although Meridian advertises these drives as 24X, they are really 32X Teac drives. We think the better performance is due to the higher speed of the TNT CD-ROM drives, with no network effect. More puzzling was that running the same copy operating from the cached drive on the TNT disk array yielded a 20% worse performance. This factor we ascribed to the parity of the striped set.
We saw transfer rates from 0.75MBps for local CD-ROM drive access to between 1.15MBps and 2.03MBps for the TNT server. We could not measure the six-fold increase that the vendor claimed for hard drive caching over running an application from the physical CD-ROM disk.
You should get better performance from a striped set without parity, so we reconfigured the drive set that way. To do this, we first removed all shared volumes, all cached drive volumes, and any programs that access the striped set. For example, Microsoft Bookshelf loads a small program that initiates at startup, which prevented our deleting the drive set in the Disk Administrator even though all shares were removed. That done, we reformatted the disk set. You do not need to unload the CD Net software to perform this task. The setting for maximum read performance or network throughput is part of the NIC card`s setup and was not changed for these measurements.
After running the same file copy, we got similar results from the copy from the TNT server`s CD-ROM drive. This was expected, since there was no change there. Our file copy from the cached CD-ROM drive on the disk array improved, becoming about 15% less (faster) than from the physical disk in the CD-ROM drive. But this was still not the great performance improvement for a cached disk that we expected to see. We did note one interesting feature: There is a very efficient caching mechanism for the CD-ROM drive in the system.
Our second copy of the same 27.9MB folder from the CD-ROM drive showed nearly a 35% improvement in speed over the first time we performed this measurement. This was almost as fast as our fastest transfer from a cached disk. We ascribed this result to the performance of the CD-ROM cache and its ability to read ahead. Typically, CD-ROM drives have the ability to cache 10% of the capacity of a disk--in this case, about 64MB.
To defeat caching effects both for the CD-ROM`s cache and the Windows NT cache, we performed a very large file transfer. We copied an entire CD-ROM disk of 638MB from both a cached disk and the same physical disk in one of the CD-ROM drives. Before running either disk copy across our 100MB Ethernet network, we rebooted the TNT Server to flush the caches. The rates we observed were 1.59MBps from the physical disk and 2.10MBps from the cached volume. Transfer of an entire CD-ROM drive from the cached disk gave us the best transfer rate from this system.
TNT comes with a performance utility that measures network throughput. We tried to use CD Net Tools to get throughput during file copies, but the amount of data transferred pegged the counter to its maximum value. We couldn`t adjust the scale high enough to accommodate our experiments. This is something Meridan will probably fix in an upcoming release.
We didn`t see a significant increase in the speed of access to the data across the network for the TNT`s 24X CD-ROM drive as opposed to the disk array when we measured sequential reads. And even though DVD drives are rated slower, an 8X DVD drive would typically deliver nearly the same data rate as a 24X CD-ROM drive.
The set of experiments described above should give the maximum performance of the system for both cached and uncached disks. We did sequential reads, and thus positioned the head of both drive types over large files. That allowed both the hard drive and the CD-ROM drive to look ahead to the next sector containing the file. High speed CD-ROM drives have sequential read rates approaching that of a hard drive (Meridian quoted a 3.5MBps maximum transfer rate for a 24X drive for sequential reads). So these transfer rates mask the potential difference between random reads of data that one might experience in the TNT`s production environment.
Randomly Reading Ratios
To test a random read situation where we might be able to measure a more realistic performance difference between the cached and uncached CD-ROMs, we did a search for all of the INF files on a CD-ROM. INF files are small (from 1KB to 120KB), so reading them shouldn`t allow the disk to read ahead. These files are also scattered about the disk.
We used the Find command on the Start menu on a Windows NT client to find INF files on a TNT Server share from across the network. Our search of a Windows 98 RC1 disk yielded 202 files. We then selected all files with unique names. When duplicate names occurred, we selected the first occurrence of that file name. That left 160 files, 1.98MB in size, to copy. This experiment should simulate the situation that occurs in a production environment for the TNT Server when multiple users are accessing the same disk and performing different tasks.
After each measurement, we rebooted the machine to avoid any caching affects. When we drag-copied these 160 files from an uncached disk, we observed a transfer rate of 0.19MBps for the uncached disk and 0.32MBps for the cached disk. This was a performance improvement for the cached disk over the uncached disk of about 170%--not the 600% advertised, but still a significant improvement in performance.
In theory, the number of connected and accessing users to a CD-ROM cached to a hard drive should be increased over the number of connected users to a CD-ROM drive, due to the faster speeds for reads in the devices. We didn`t test this loading effect, but in a random-read situation it would make sense, given the better performance of a hard drive over a CD-ROM drive.
We wondered how the caching function of the hard drive and CD-ROM would affect the performance of our random-read experiment. Copying the same file set to another folder on the same drive on our client from the TNT Server dramatically improved the performance in both cases. For the CD-ROM drive, we measured a rate of 0.61MBps, or a 320% increase. For the cached disk on the hard drive, our second read utilizing the cache gave a rate of 0.67MBps, an improvement of 207%. Taking caching into account, the cached hard drive was only 10% faster than the uncached volume in the CD-ROM for random reads. However, we think that in a multi-user situation the effect of the cache would be limited.
Regardless of the performance level, the library function of the TNT Server is valuable in an enterprise. The TNT Server`s ability to maintain large numbers of stored CD-ROMs, to use cheap disk space to store CD-ROMs and give network access, and to work with jukeboxes are major advantages over devices that are simply CD-ROM carousels or jukeboxes serviced by CD-ROM drives alone.
Nothing but Net
From the user`s point of view, using the TNT Server is seamless, once the software and volumes are properly set up. CD Net lets you have applications reside on a users desktop or Start menu, or it offers a small GUI application called CD User that you can provide to clients. When users open CD User, they select a database of applications, and each application runs from an icon in the window of that utility.
Typically, an administrator creates a database of installation applications for CD-ROMs specifying the SETUP.EXE file. After clients install the application on their local hard drives, they open another database that will launch the application from an icon. The administrator may have to modify the target and working directories for the shortcuts to applications in the CD User database that runs those applications in instances where they aren`t UNC compliant. This aspect of setup can be confusing and could be made easier.
Another method for presentation, TNT Server`s Slingshot! software (a set of CGI programs) can work with IIS or Netscape Fast Track Server to present web pages that users can browse to launch applications or to view the contents of a volume. Meridian has some standard pages that are easily modified with your own information.
Product: Meridian TNT Server, Tower 14
Company: Meridian Data
Hardware includes seven 24X CD-ROM drives; one DVD-ROM drive, four 6GB hard drives, 233MHz CPU, 128MB RAM, one floppy disk drive, and 3COM NIC. Software includes Windows NT Server 4 SP3, IIS, IE 3.02, CD Net V4.0 for Windows NT, and CD Intranet software.
The Meridian TNT Server would be valuable to any enterprise where there?s a need to create and manage an on-line library of data, documents, or applications across a network. A key feature is the ability to cache and manage volumes to a hard drive, improving both speed of access and number of connected users. The software can be confusing to work with at first, but has the necessary feature set.
This article is reprinted with permission from BackOffice Magazine, a sister publication of InfoStor (www.backofficemag.com).
CD Manager automatically recognizes disks inserted into your drives if you configured that option in setup. Using this utility, you can create volume groups (1), which are the key to creating shared groups of disks and managing security. Shared volumes appear as green icons (2). Cached volumes (3) have a small hard drive icon attached to them.
This chart shows the results when 160 small files were read from the physical drive and the cached drive. Without caching effects, we observed slow transfer rates and a 170% faster transfer from a cached disk than from a physical disk. When CD-ROM and system caching came into play, the transfer rates doubled or tripled and the differences between the devices disappeared. For environments where lots of small, separate files are read (like help systems) the cache offers better performance. For large files or contiguous files, as there would be in installation disks, there is little difference between the physical disk and the cached disk.
This chart shows the effects of streaming file sizes large enough so that the read heads can "look ahead" to the next sector. Network reads over a 100MB Ethernet network using TCP/IP were faster than those of a local 12X drive. Speeds increase when you go from a striped set with parity to a striped set without parity, due to the loss of parity checking. Our best transfer rates were achieved when we copied an entire CD-ROM volume (638MB) and allowed caching to come into play.
This chart shows the difference between transferring an entire CD-ROM over the network rather than small file copies. Streaming file copies are much faster than essentially random copies. When we copied the same set of files again and let the CD-ROM and system cache come into play, we saw performance enhancements in both instances, but streaming file transfers were still much faster.