Server Or Workstation – Pick Your LAN Backup Antidote

svrorwsDespite claims by various backup-system vendors about which approach to network backup is better — workstation- or server-based — tests of both implementations show that the server-based solution has some glaring disadvantages. However, because there’s only one top-notch workstation-based implementation currently available, network administrators are in a catch-22 when it comes to making a good buying decision.

Server-based vs. workstation-based network backup: The holy war over which method is better has gone on for quite some time. After a year of testing virtually all of the available contenders in each of these categories, we are better able to put the advantages and disadvantages of these two methods in perspective.

Now, with Palindrome Corp.’s introduction of its NetWare Loadable Module (NLM), there is at least one vendor with a foot on both sides. Representatives from both corners are very quick to point out the advantages to their approach, and the disadvantages of the opposite corner’s.

Workstation-based implementations, or “tape stations,” are not an integrated part of the server. Therefore, the actual backup process cannot interfere with the performance or stability of the network operating system’s core processes.

On the other hand, tape stations must be located across the network where, once invoked, the backup process may have a negative impact on overall network performance. In addition, security-conscious installations may have reservations about leaving tape stations unattended while sensitive data is being backed up.

In contrast, server-based solutions are secure because the entire backup process is internal to the server. Compared to the tape station approach, the server-based approach is usually faster because data transfers occur across the server’s bus instead of across the network.

However, provided that there is more than one file server, or individual workstations with hard drives that need to be backed up, the performance advantage is only applicable to the one file server in which the backup-software process is running. The rest of the nodes must send their data across the network.

In the last year, we has reviewed several tape-backup solutions of both the server-based and tape station-based kind.

Which type of backup solution is better? The decision to buy depends on the needs of an organization and the willingness of the network administrator to put up with the disadvantages of each type.

However, after a year of testing these products, we have found more disadvantages in implementing and testing the server-based approach.

In the server-based approach, the software that controls the backup process must be executed as a separate task on the file server. In the case of Novell Inc.’s NetWare, which most of these products support, this means that the server must run an FSP, short for File Server Process. For NetWare 3.11 servers, an NLM is required. For NetWare 2.x servers, a Value-Added Process (VAP) is necessary. The tape drive must also be attached to the same server on which the FSP is running.

The major advantage of this architecture is that the transfer of data is self-contained within the bus of the file server. The data moving from disk to tape never makes it onto the network wire — therefore, performance is not affected by the potential bottleneck of that wire and there is no effect on the existing network traffic.

The server-based architecture also allows for the easy incorporation of the client/server model into the design of end-user software. Workstations can easily submit “jobs” or backup requests by updating a file on the server, which is monitored by the FSP.

Currently, however, the disadvantages to the server-based method outweigh the advantages.

For starters, installing a server-based system invariably requires some sort of SCSI (Small Computer System Interface)-based support for the tape drive. If the current server’s current drive subsystem is not SCSI-based, some dismantling and reassembly of the file server may be necessary.

Balancing Act

While adding SCSI cards to a file server is not a problem, there exists a delicate balance between the SCSI firmware revisions, the network operating system drivers for the SCSI interface, and the backup FSP.

Synchronizing all the drivers for compatibility can be extremely difficult. For example, there are countless numbers of SCSI-driver disks with Adaptec labels on them; some with the same revision levels on the label, but with differently timed stamped files. Other labels indicate new revision levels, but have older dates than the ones they’re supposed to replace. The one that works for one system may not work in another, and chances are that neither were tested by the tape-backup vendor.

In computers that already have a SCSI subsystem installed, that same delicate balance is already in place for the disk subsystem instead of the backup FSP. As evidenced by testing, using the same SCSI channel for both is a dangerous game. The same drivers that work perfectly for our disk drives do not always work for our backup system.

All too often, this tangled web of drivers generates something that many NetWare administrators are familiar with — a General Protection Processor Exception. With the exception of Cheyenne Software Inc.’s ARCserve, every server-based backup tested by PC Week Labs eventually and inconsistently crashed a server.

Proponents of server-based backup may argue that the same driver-synchronicity problems can exist in a tape station scenario. However, when tape stations are used, the reliability of the file server isn’t affected — an important consideration if mission-critical applications depend on that server.

Server-Based Supporters

Server-based pundits will also argue that server-based backup offers clear performance advantages since the data never enters the network — a requirement when using tape stations. However, the major advantage of network-based backup systems is that they can back up all network nodes from one location.

Unless the network-based backup system was installed into one file server, and was meant solely for the purpose of backing up that file server and nothing else on the network (an implementation that eliminates the need for network-based backup), network bandwidth will eventually be affected anyway.

Besides, at least one vendor — Gigatrend Inc. — has proven that the tape station approach can be done without compromising performance.

In tests performed at our labs and at the sites of St. Agnes Medical Center, in Fresno, Calif., Gigatrend’s MasterDAT proved capable of moving data from the server, across the network to the tape station at speeds that matched or bettered some server-based solutions.

Perhaps one of the most compelling reasons to take a second look at the tape station approach is the process of recovering a crashed server volume.

In addition to the backup software itself, most server-based backup systems store their control data (indexes, pointers, etc.) on the server. If the server volume that holds all of this vital information crashes, the recovery process is a bit more trying on the network administrator’s nerves.

First, the volume must be restored to working order, a step that applies to either server- or tape station-based approaches. Then, the backup software must be reinstalled to that volume from the original installation disks; the indexes and control files must also be restored. As some backup systems are incapable of backing up files that remain open during the backup process (namely the systems’ own pointers and indexes), restoring the control files may require rebuilding the entire tape catalog. This proce ss could take several hours, since the computer must read every tape from beginning to end. Once all the control files have been restored, the volume’s data can finally be restored.

As an alternative, the tape station approach should take no fewer than two steps: restoration of the volume to working order, and restoration of the data.

Finally, today’s networks are heterogeneous. They have NetWare servers, Banyan System Inc.’s VINES servers, LAN Manager servers, Macintoshes that talk AppleTalk, Unix boxes that talk TCP/IP, etc. Currently, there is no network-based backup product that communicates to all of these node types because no one network operating system provides all of the different protocols needed to communicate simultaneously with a variety of nodes.

Despite trends in that direction, putting the burden of communicating with all of those different node types on a file server, for the sole purpose of backing up, adds an undesirable a level of complexity to the server’s configuration. Doing so may ultimately have the same impact as the imbalance between SCSI drivers and FSPs discussed earlier.

File-Server Burden

Entrusting a file server, which a company depends on for mission-critical applications, with the responsibility of backing up the entire network is not an attractive proposition. Of course, with products such as Novell’s NetWare Runtime, one could ignore the mission-critical aspect and dedicate one server to network backup — in this case it’s no longer a file server, its a really a tape station.

With the exception of AppleTalk, Anasil has been successful in loading all of the key protocols on one DOS workstation simultaneously. Considering this, the tape station approach looks more attractive since it is currently subject to fewer communications limitations.

In theory, the tape station approach is better. The technology to communicate with an entire network is here today, and it puts the backup system at a comfortable distance from a network’s most critical resources: the file servers. Also, there is no reason why the client/server model cannot be applied to the tape station approach.

In practice though, there is only one tape station-based system that is worth looking at — the non-NLM version of Palindrome’s The Network Archivist. Gigatrend offers superior performance, but the recovery process is still somewhat kludgy.

The server-based contenders, such as Emeritus Technologies’ TapeWare and Legato Systems Inc.’s NetWorker, are better at what they do, and are not nearly as cost-prohibitive as the Palindrome offering.

As a result, network administrators are currently in a catch-22. Until some more first-rate offerings are available in the tape station approach, the war will continue.

Leave a Reply

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

  • « Older Entries
  • Newer Entries »