Ask Dr. Bob Your NT Questions - 01 Feb 1997

Send us your tips and questions. You can also visit Bob Chronister'sonline Tricks & Traps at http://www.winntmag.com/forums/index.html.

Q: I've seen many discussions about the use of symmetric multiprocessing(SMP) in Windows NT. Many users favor it, but many say that it's actually slowerthan using one processor. Which is correct?

Both sides are correct. Many factors enter into the use of SMP. Mostsystems use the Intel SMP configuration (MPS 1.4), but many use OEM-specifichardware abstraction layers (HALs) optimized for specific motherboards. Thelatter systems tend to work better than the generic MPS 1.4 HALs. In the race tobe different, a board manufacturer will often add or change features that arenot compatible with the MPS 1.4 specification. These changes can diminish theSMP performance.

In both standard and OEM-equipped SMP machines, the ultimate performancedepends on which application you're using. SMP balances a load symmetrically(i.e., across all processors). The specific thread characteristics of theapplication affect the load balancing. If the application is not thread aware(which usually means the vender never tested the application on an SMP machine),an SMP machine will not perform as well as a single-processor machine. Screen 1shows an application (U-Lead's Photoimpact SE version 3.0) that is poorlythreaded. Notice that the workload does not appear to be symmetrical; in fact,the CPUs appear to be antagonistic toward one another (i.e., the thread bouncesfrom one CPU to the other). In contrast, Screen 2 shows an application (PicturePublisher for NT, which unfortunately was never developed into a full product)that is very thread aware. The CPUs are in total synchronization and completethe work much faster than a single-processor machine.

Caution: A friend recently purchased a dual 200MHz Pentium Pro witha SuperMicro motherboard. He added an Intel Pro 10/100 NIC and decided torepartition his hard disk. NT saw only one CPU. When he removed the Intel NICand installed a 3Com XL card, the system recognized both CPUs. Be aware thatcertain NICs can conflict with the Advanced Programming Interrupt Controller(APIC) logic in the SMP machines.

Q: How do image backups compare to standard file backups? Is this a wayfor backup vendors to get us to buy new software?

Mainframes have used image backups for many years, but file backups aremore recent. In general, an image backup works with the sectors on a disk and isindependent of the sector content. An image backup includes information aboutpartition tables, file tables (FAT, MFT, etc.), and the Master Boot Record. Filebackups contain information about files and their attributes. In a file backup,you can selectively restore individual files, whereas in an image or sectorbackup, you have to basically restore the entire drive.

Both types of backups have advantages. The file backup lets you restoreindividual files and directories but will not regenerate a drive. The imagebackup lets you boot to a set of floppies and restore the contents of a tape,thus regenerating your drive. However, the size of the drive you restore to mustbe at least equal to the size of the drive that you back up. In fact, an imagebackup works best when you use identical drives. In all cases, you need toperform a low-level format on the drives to optimize the restore. You do notneed to partition or format the recipient drive. As you might expect, you can'tdo an incremental sector backup, so image backup is better for true disasterrecovery and file backups are better for individual file restoration. As of thiswriting, only ARCserve and ULTRABAC offer both image and file backup softwarefor NT.

Q: My colleague wants to use a print server on a small NT LAN. He claimshe can set up the print server as a TCP/IP device, which will offer greatadvantages to the LAN. What is a print server, and how do you configure it as aTCP/IP device?

Print servers are small devices that connect directly to a network hub orswitch. I am aware of only servers that function with 10Base-T and 10Base-2connections. The print server I have the most experience with is the Axis 560.This device has two parallel connections and one serial connection. You simplypower the device after you connect it to the switch or hub. From this point, thesetup becomes an NT and print-server issue. On the NT side, you need to addMicrosoft TCP/IP Printing under the Services tab of the Network applet and FTPunder the Internet Information Server Installation applet.

Step one: Place a printer on the first parallel port (PR1), and print a testpage. This page will most likely give you the printer's current settings. Theone to notice is the IP address, which is 192.36.253.80. Unless you have arouter or some means of accessing the device, you need to change the device's IPaddress.

Step two: Decide which machine will control the print server. The BackupDomain Controller (BDC) is a logical candidate. You can use a workstation forthe print server, but this approach limits you to only 10 simultaneousconnections. Go to the Network applet and change to the IP address of thedesignated machine. This step is not a major issue because you won't have toreboot. Within the subnet of the print server's IP address, choose an addresssuch as 192.36.253.82. After you apply this new address, type

PING 192.36.253.80

at the command prompt and press Enter. Make sure you can ping the printserver before continuing to step three.

Step three: At the command prompt, type

FTP 192.36.253.80

and press Enter. You will connect, as you see in Screen 3, and receive aprompt for user name and password. The default settings for Axisprint servers are root (as user) and pass (as password). Afteryou establish a connection, you need to access the configuration file. Type

get config

and press Enter. The configuration file, config, will be in the rootdirectory.

Step four: Open config in Notepad, as you see in Screen 4. Change to aunique IP address on the print server's subnet by changing the address listed,and save the config file. Upload the config file to the print server by typing

put config CONFIG

and pressing Enter, as Screen 3 illustrates. Print a new test page. The newtest page will show the new IP address. Use the Network applet under the ControlPanel to return the machine's IP address to its original address.

Step five: Create a new printer on the machine that controls the printserver. This printer will be a local printer. Using the standard add printerwizard, add a new LPR port, which you see in Screen 5. When you see the promptto enter the name or IP address of the print server, enter the IP address youassigned the server. For the name of the printer, enter the print port or theprint server where the printer is connected (e.g., PR1 or PR2). Choose theprinter type and load the appropriate driver. You then simply share the printeras a standard printer share. Screen 6 shows the new PR port highlighted.

The server will work well on the network. However, be aware that anyprinting on the network will tie up the controlling machine. Each print requestwill cause major CPU utilization, as you see in Screen 7. For occasional printjobs, you can specify the controller as any machine on the network. For aserious print server, dedicate a machine.

Q: I have a question about the Queue Length counter that you can enable inNT's Performance Monitor. Do you have to monitor at least one thread counter tokeep the counter from always showing 0?

The System:Processor Queue Length counter is an important tool formonitoring excessive threads. A lot of multitasking can generate an excessiveamount of threads and produce a bottleneck. In NT 3.5, you must monitor at leastone thread for the counter not to read 0, but in NT 3.51 and 4.0, you do notneed to monitor a thread.

Q: I am trying to do unattended installations, but I can't figure out howto turn off the End User License Agreement (EULA). If I can't turn it off, Ican't perform unattended installations in NT 4.0.

Believe it or not, you can turn off the EULA. However, you need to be awareof the consequences. Microsoft wants everyone to see the EULA and agree to it,whether they read it or not. If you turn the agreement off, you are violatingMicrosoft's intent. This action is not a major issue because you can print outthe EULA and give the hard copy to the machine's user. To turn the EULA offduring unattended installations, you need to add the following statement in theunattended section of your answer file (unattend.txt by default):

[Unattended]

OEMSkipEULA=yes

Q: I am interested in adding modems and ports to our company server.Unfortunately, the advertising for such products is very confusing. How do Iknow what type of device to buy or know what is available for use?

You can take at least three different approaches, from the simple to thepowerful.

1. Add a Standard Multiple-COM Port Card

Several vendors, including Boca Research and STB Systems, make multi-COMport cards. If you have enough open IRQs, these cards work well. However, beaware that communications servers are generally resource intensive, and theylack some serious communications server features: For example, they don't reduceCPU workload.

2. Add Intelligent Multiple COM Port Cards withOnboard Processors

An intelligent multi-COM port card with an onboard processor can reduce CPUworkload. An example of such a card is the Digi International PC/2e board, whichyou add in the Network applet in Control Panel. Once you assign a memory baseand I/O base and reboot the system, the card is ready to use. Screen 8 shows anexample of a successful configuration with two COM ports available. Using thesecards with full-sized modems or with numerous modems can easily become ahardware nightmare because of the size of the setup and the amount of cablingrequired.

3. Add Modem Servers That Use PC Card Modems

Modem servers with PC Card modems are small and easy to manage. Two types ofthese servers exist. The first type is a network server that uses either TCP/IPor IPX/SPX to attach to a network, and the second type is a SCSI device thatworks off a special driver.

Digi's LANAserver 6e is an example of the first type of modem server. Thisserver has six COM ports and is easy to configure. I set up one to work withTCP/IP. You simply run an onboard main menu and add the IP address andsubnet mask. The board then resets, and you can easily PING it. Install theLANAserver application and configure the server on the network. Interestingly,the program adds a layer of security to the system--you must pass the serversecurity to gain access to NT's built-in security. Screen 9 shows the securitywindow.

After you set up the system, all modems are visible in the LANAserverManager, as you see in Screen 10. You can then edit each port for type of card(LANAserver can use Integrated Services Digital Network--ISDN). The modemserver's design is very impressive, and because it is a peripheral server, itconsiderably reduces the workload from system CPU.

Despite these promising features, Digi does have a few problems to workthrough. For example, the server expects to load IPX/SPX. If IPX is loaded, youcan dial out from any node on the network. Unfortunately, you can't dial outwith NT, but Digi assured me that NT support is in the works.

Another type of server is Central Data's equally elegant SCSI-based ST-1008+communications server. You assign this server an unoccupied SCSI ID and connectit to a SCSI controller (I used an Adaptec 3940 with firmware 1.14). After youreboot the system, you load a driver through an easy-to-follow standard setup.Screen 11 shows the Central Data driver loaded in the Devices applet under theControl Panel. Unlike Digi's LANAserver, Central Data's communications serverconsiders each of its ports as a standard COM port and lets you configure theserver through dial-up networking or Remote Access Service (RAS). Screen 12shows the eight PC Card slots added as ports in the Port applet under theControl Panel. (As a side note, Central Data also offers IP-based servers.)

The obvious advantage of Central Data's ST-1008+ server is the use of a SCSIbusmaster to off-load CPU work. It is an elegant approach to a communicationsserver.

All these devices have clear-cut advantages and disadvantages. The four-portcard is inexpensive but CPU-intensive. For moderate access, thecoprocessed cards are excellent choices. For intensive use, the Central Data andDigi servers are well worth consideration.

Q: I recently purchased a 3Com Impact IQ ISDN device. All the instructionsfocus on NT 3.51. How do I configure it for NT 4.0?

Installing your 3COM ISDN device for NT 4.0 is simple. Ignore everythingabout setting up the device with NT 3.51 RAS. Go to the Modems applet under theControl Panel. Click on the button to add a new modem, and tell the system thatyou will select the device from the list (i.e., do not autodetect the device).Choose Have Disk, and insert the 3Com driver floppy in the drive. Whenthe system presents you with a list of 3Com modems, select the 230 (even thoughmost Internet Service Providers--ISPs--don't support compression now, they mayin the future). The rest of the process is automatic. Reboot the system. Theonly caveat to using any ISDN device on a serial port is the ability of thatport to support high-speed connections. I strongly recommend a Digi board (oranother supported coprocessed card) for the connection.

Q: Like all systems administrators, I'm concerned about protecting systemsfrom data loss and failure. I know of one system that got a boot sector virus,and the whole drive was lost when the boot sector was repaired. Can I protect mysystems from this type of situation?

I've been working on this issue for quite some time. First, this situationis not so grim if you're prepared for it. You can easily recover from this typeof data loss if you have a simple set of tools.

1. Always have an up-to-date Emergency Repair Disk handy. Always runrdisk /s after major changes to the system.

2. Always have a recent backup available--not having one is almostcomputer malpractice. Backup applications are becoming cheap and plentiful. Irecently looked at McAfee's QuickBackup. Although it ostensibly functions in NT4.0, you can back up to a remote FTP server or to a hard drive, but a backup totape won't work. Interestingly, the cost of the program is less than $50. Nowyou have no excuse for not having a backup handy.

3. Make an NT boot floppy. Most people think a boot floppy is onethat allows only DOS booting, but you can also have an NT boot floppy. It issimilar to a DOS boot floppy, but easier to make. The beauty is that you canmake an NT boot floppy on any NT machine (if you are on the same platform). Theonly variable is boot.ini.

Format a 3.5" floppy in NT (File Manager is still the easiestapplication way). Copy ntldr and ntdetect.com to the floppy. I usually make sureI'm using standard file attributes before copying these files because standardattributes make the files easier to work with. Copy boot.ini to the disk. ntldrand ntdetect.com can come from any machine, but boot.ini is machinespecific (so make sure you have a copy of boot.ini handy to copy to the floppy).Don't worry about copying bootsect.dos. Because obtaining a DOS boot floppy iseasy and bootsect.dos is machine specific, use a DOS boot floppy.

4. Every serious NT machine needs at least two hard disks. Pick oneas your main disk and one as a secondary disk. Place a copy of NT on each disk.Although it sounds like work, this configuration can be a godsend. If theprimary NT system fails, you can always boot to the secondary one to repair theprimary one.

5. Make sure you have the NT boot floppies. If you don't, run eitherwinnt /ox or winnt32 /ox (you can run the first from DOS or Windows 95, thesecond from NT, or both from the appropriate folder on the NT CD-ROM).

After you finish these five steps, you're prepared for nearly any disaster.Let's look at two examples.

You Have a Corrupt Boot Sector

Boot to the NT boot floppy, and you will have the same NT options you hadbefore the corruption. You can gain access to critical files and make a backupif necessary. After you've taken these steps, you can try repairing the bootsector. If the repair does not succeed, you still have saved all the criticaldata.

You Get an Event Viewer Message That a Hard DiskIs Failing

Back up the failing disk. Turn the system off, remove the bad disk, andreplace it with a new one. Insert the boot floppy, and boot to the copy of NT onthe remaining disk. You can open Disk Administrator and format the new disk andthen restore files.

Screen 13 shows a normal setup. Disk 0 has three partitions. NT 4.0 wasinstalled on Drive E and Drive H. Disk 0 was removed and replaced with a blankdrive. The system was booted to NT on Drive H. Screen 14 shows Disk 0 as beingblank. This situation is easy to fix. Restore the backup of the lost disk onto anew one. Boot to the first installation floppy, and run a repair of the bootsector. This total process will take about 30 minutes depending on the size ofthe drive. (Alternatively, you can back up and restore a sector. This method isfaster but not necessarily as up to date.