|
Contents:
Server Based Vs. Wire Based Network Backup System
There are two main ways to backup file server or workstation hard drives. The first is to attach a tape drive directly to the file server or workstation. This is the fastest way to backup a file server, but it will also require you to purchase a backup unit for every system needing backup.
The second is sharing backup devices across the local area network (LAN) wire. This allows one tape drive to backup multiple file servers or workstations. This alternative is less costly but it can also be somewhat slower since all the data being backed up will have to go across the wire.
There are currently many network backup solutions available; including server based tape systems such as Cheyenne's Arcserve and Palindrome's The Network Archivist, workstation based solutions such as Gigatrend's MasterDat and Mountain's Filesafe, and the recently introduced dedicated backup server, Intel's StorageExpressTM System. All of these solutions will back up servers and most will back up work-stations. The server based systems are attached directly to a file server, which they back up directly, and the other file servers and work-stations on the network can be backed up across the wire. Workstation based systems require a dedicated workstation during the backup, and back up file servers and workstations across the wire.
Intel's dedicated backup server also backs up all file servers and workstations across the wire, but does not require a dedicated PC because all of the intelligence needed is built into the systems hardware and software. The main advantage of a dedicated backup server is that it will not use up resources on the file server.
The Intel StorageExpress system's high level of integration and intelligent software make it easy to install and very reliable. The StorageExpress system has also been tested and certified as a complete system, in contrast to alternative approaches which all require integration by the customer. Many times the backup software and hardware have not been fully tested with a wide variety of file server platforms.
Alternative approaches can also have hidden costs such as additional memory, a larger hard drive, or a faster CPU. A server based tape unit uses memory and other file server resources to operate. Both the workstation and server based systems require the end user to down the file servers and add all of the hardware and software components to their network. They must also deal with any incompatibilities or NLM conflicts that arise and ensure that all the pieces work together. Overall Network Performance
Many things can effect the performance of backing up data to a tape. Some will have more effect on overall performance then others. Since there are several types of backup solutions, some of these performance issues may not effect all types.
Let's first take a look at a network. This could be a local area network
or a wide area network. Obviously, if large amounts of data are to travel around the network they should go the fastest route possible. This way, the most amount of data can be backed up in the shortest time. Network Topology
Network topology can have a large impact on backup performance. Ethernet allows data transmission at 10Mbps, and Token Ring at either 4Mbps or 16Mbps. Ethernet uses a carrier sense, multiple access, collision detect method; therefore, with more network traffic it can reach a saturation level. Token Ring passes the data from one workstation to another, so that the network can grow larger without reaching a saturation level, but it will take longer for data to go completely around the network ring segment.
More important is the amount of data that can be sent in each packet. In the Novell NetWare environment, Ethernet supports 1Kbytes of data per packet vs. 4Kbytes of data per packet for Token Ring. So if a 1Mb file needed to be backed up over the wire, Token Ring will be able to send the data in fewer packets. Another topology that is supported by Novell is Arcnet, which has a maximum data packet size of 512bytes. Bridges And Routers
Bridges and routers will generally slow down large amounts of data moving through the network. These devices are designed to segment networks and not used for moving high amounts of data; this includes the use of Novell internal routing.
A good rule of thumb with any backup solution, is to keep the tape backup system on the same wire segment or ring on which the devices being backed up are located. This may mean a redesign of some of the network in order to backup efficiently, or the purchase of additional backup units.
Even though some tape systems can be managed remotely, you should try to avoid having the data to be backed up going across a bridge or router. The mega-bytes per minute rate will drop even further if you go across more than one bridge or router.
Your Novell file servers console (display servers), or the Intel NetSightTM Sentry monitor can help determine your network design. If it is necessary for data to cross a bridge or router, make sure it is the fastest bridge or router affordable.
When using Novell's internal routing, use NetWare 3.X on a high speed PC. Keep in mind that Novell internal routing is not designed for high speed routing, although Novell does have products based on NetWare Runtime that will provide much better performance over internal NetWare routing.
A problem with data transferred across a router is the way NetWare handles routed packets. When a workstation logs into a local file server a packet negotiation takes place. The workstation and the file server will negotiate to the largest packet size. In Token Ring the largest can be 4Kbytes of data per packet, with some overhead for source and
destination addressing and general error checking.
In a NetWare* routed packet, the file server does not know the data source so the lowest common denominator is chosen, 512 bytes (Arcnet). Novell recently released a product called BIGPACK.NLM; this is an NLM that will allow NetWare file servers to negotiate routed packets to higher levels.
When two NetWare* file servers communicate, they do not go through the same packet negotiation as when a work-station talks to a NetWare file server. The communications between the two NetWare file servers will always default to 512 byte packets. This means that when a file server based backup system attempts to backup another file system across the wire it will always be using 512 bytes of data per packet. Novell has recently announced a new CLIB.NLM, which, when loaded on a file server or backup server like the Intel StorageExpress™ system, will allow packet negotiation. In testing it has been found to yield up to 40% faster performance than today's CLIB.NLM when doing backup. How To Measure Network Backup Performance
Backing up network file servers and workstations requires moving from a hard drive to a secondary type of media such as magnetic tape. That data can be moved in several ways, such as attaching a secondary storage device directly to the system being backed up or backing up across the network or wire. Both ways are totally acceptable as long as the data is moved in timely manner and will be free of errors if needed for a recovery.
Generally, backup performance is measured in mega-bytes of data copied to the secondary device per minute. For this paper we will be using tape backup as our secondary storage medium, although it could also be a WORM drive, a magneto-optical drive, a juke-box or another hard disk. Network Wire Load
Network wire utilization is where network backup performance suffers the most. Since wire based backups require the transfer of data across the wire, it's a good idea to backup during very low wire load times.
Wire utilization can be as high as 10% to 15% during backup, and higher when doing multiple backups simultaneously. Many systems support unattended backups to do the backups late at night when network traffic is low and most files are closed. In large LANs it is also nice to have a backup system that can help determine the overlap of backups to allow maximum use of the wire's bandwidth. The Intel StorageExpress system's Central Console software provides a backup histogram that shows the currently scheduled backups, so that a backup administrator can take full advantage of the wire and won't overlap backups. Noisy Cabling
In Ethernet and Token Ring topologies, data can be run on both shielded and unshielded cable. Shielded cable, though more expensive, can handle
environmental disturbances much better than unshielded. This means that if the media is located in a noisy environment there will be many more errors and retries on unshielded cable. Environments that generate errors include, power sources, improperly grounded cables, lights or basically anything with AC power. How do you find if the cable has errors or retries? Use a network analyzing tool like the Intel NetSight Professional or Sentry analyzer or monitor These products will identify errors or retries on the media.
In summary, there are many things about the overall network design that can enhance or degrade the backup performance. The use of Ethernet or Token Ring, the choice and placement of routers and bridges, and the size of the packets going across the network. Take a good look at the network layout before attaching any backup device and start sending large amounts of data to be backed up. File Servers And Network Backup Performance
The file server or workstation being backed up can also have a large impact on backup performance. The faster the data can be removed from a system and sent across the wire, the larger the mega-byte per minute backup rate will be. What can effect this performance? Hard Disk Drives
First of all, the data is stored on a hard disk. Most disks have an access time, ranging from 10-40ms. That number is the amount of time it takes to retrieve data from the hard drive. By using technology like SCSI2 controllers and SCSI hard drives, that number can drop toward the lower end of the range Check with the disk controller and disk drive manufacture to find out the access time of the disk system in the file server or workstation. Network Interface Cards
The type of the network interface card (NIC) installed in the system can also greatly affect overall performance. Don't put a slow NIC into a high-speed file server and expect great performance. Make sure, especially in the file server, that the fastest card and the fastest hard disk are installed.
Most NIC manufactures provide performance numbers for their cards. Not only is NIC throughput performance critical, but also server utilization. The NIC may be very fast, but may take up large amounts of CPU utilization. A busmaster type card may have a little slower through-put, but will require much less CPU utilization. CPU utilization is critical when doing a backup while other activities co-exist on the file server.
Testing different NICs with different types of SCSI devices and different buses has shown that an integrated NIC and integrated SCSI2 performed the best. This configuration is used in the StorageExpress system to maximize throughput.
File sizes and disk fragmentation can also effect the performance of backup. In testing it was found that if a hard disk with 30MB of data
was backed up, 8 large files totaling 30MB backed up faster then 1000 files totaling 30MB. Also, if those 1000 files exist in the same directory (at the same level), the backup was slower than if the 1000 files were dispersed with each directory containing less than 20 files. System Memory
How does the amount of RAM in a system affect the performance? There are two places for RAM to affect the backup performance, in the machine being backed up, and in the system that is doing the backup.
In a server based backup, the RAM is shared between the file server processes and the backup process. Most file server based backup systems require adding additional amounts of memory to the file server that has the backup device installed.
The Intel StorageExpress™ system has its own dedicated 8MB of RAM (default value) which was found to be the optimum amount needed while testing the system. When additional drives were added and dual streaming was used, there was no significant performance difference between 8MB and 16MB of RAM. Remember that the StorageExpress system is a dedicated backup server. This means that it is only doing backup, and not providing any other network services such as printing or user file access. Storage Management Services SMS
Novell has recently come out with a new backup specification called Storage Management Services or SMS. SMS allows transparent backup of all Novell servers and workstations. This will be an NLM on NetWare* 3.X and a TSR for user workstations. Novell NetWare 3.X provides support for many name spaces, including DOS, OS/2, NFS, Macintosh and FTAM today. If a backup system supports SMS it will be able to backup all the name spaces as well as have easy migration between the different versions of NetWare.
The StorageExpress™ system supports the SMS specification. SMS greatly enhances the features of NetWare backup but it comes with some overhead. If the server does not use name spaces other then DOS or Macintosh, SMS is not required. The StorageExpress system will be able to automatically check to see if SMS is in use. If not, then the backup will be done in non-SMS mode, resulting in the best performance. Performance Issues During The Backup Process
During the backup procedure there are several variables that can affect performance. Does the tape system provide dual or multiple streaming? Multiple streaming allows multiple backups to occur at the same time. If there are two tape drives installed into a backup system which supports multiple streaming, then two separate backup jobs can run at the same time. The Intel StorageExpress system supports dual streaming when two drives are installed. Dual streaming will boost performance by as much a 80%.
Dual streaming also increases the CPU utilization significantly. However, since the StorageExpress system is dedicated to backup, the
increased CPU utilization will not affect any other processes as would a server attached backup system. Backup Data Verification
Data verification or data compare features can check the data backed up to tape against the information stored on the disk to ensure it's correct. After a backup, the tape is rewound and compared against the original data. This provides a confirmation of error free backup but also doubles the time to do the backup. The compare or verify can be turned off for speed but should be weighted against the lower level of assurance of data integrity. The Intel StorageExpress System
There are many issues that can improve the performance of backup; design of the network, carefully choosing network components and using a good backup solution. The Intel StorageExpress system is a centrally managed, dedicated backup server, designed specifically for multi-server network backup. It's highly integrated and optimized for backup performance using an Intel 486TM CPU and integrated onboard 32-bit Ethernet interface and SCSI2 controller. The StorageExpress system can also support 16/4 Mbps Token Ring. And, by adding a second tape drive, the StorageExpress system provides expandable capacity and supports dual streaming.
The StorageExpress™ system is a backup server which overcomes many of the performance problems which have traditionally plagued wire-based approaches, while at the same time eliminating the reliability and installation shortcomings of server based solutions.
This applies to:
| 
