REVIEWS
Bypassing The Back-Breaking Burden Of Backups: Unix-Based Solutions
by Aaron C. Young
How do you back up your data? How many backup tapes do you change each day?
How many times did your backups fail this week? If you're backing up corporate
data, these questions need good, firm answers. No one wants the pressure
of being unsure about how reliable their backups are.
The idea of backing up and restoring your data is simple: Get it on the
tape, get it off the tape. What it takes to make these two steps happen
is the hard part. Important considerations: how much data you'll be backing
up, how much you mind changing tapes, whether you need a software package
that provides for all of your Unix platforms and
possibly non-Unix platforms.
An effective Unix backup package should be easy to install and take the
Unix out of Unix as much as possible. Yet, it should still retain the power
and flexibility that hard-core administrators demand. If it isn't easy to
install and its documentation isn't much help, good technical support must
be available.
The solution must support a wide range of tape hardware, be easy to configure
and provide support for many different media types. Restores and backups
must be easily initiated and backup progress information should be viewable
as operations occur. Your backup software should be able to back up more
than just Unix clients, it should be easy to add a library, and it should
have good tape and indices manag
ement. Finally, performance should fully
utilize your network bandwidth and closely follow media hardware speeds.
Backing Them Up Against the Wall
We evaluated six Solaris-based solutions.
Each provide support for an Exabyte EXB-440 ta
pe library. Legato's NetWorker
came out on top. It offers great performance and doesn't lock you into a
proprietary tape format. Legato also provides support for the greatest number
of Unix and non-Unix platforms.
Ranked behind Legato is PDC BudTool. It doesn't use any special techniques
to accomplish backups, such as Legato, Spectra Logic or Cheyenne, but it
does the job using standard Unix tools and does the job well. The PDC server
was just a bit harder to install and get running than Cheyenne's ARCserve/Open,
but it was just about as easy as NetWorker and Alexandria. Its true colors
are brought out in performance and its use of standard Unix tools.
We were very impressed with NetWorker and BudTool. Although very different
packages, they both provide an administrator with the ability to perform
a fast and manageable backup. NetWorker should be your overall choice, especially
if you're security minded. BudTool is a great alternative if you need to
keep your costs per client down.
We found Spectra Logic's Alexandria lacking in performance and it could
benefit from a simpler and easier-to-use interface. Cheyenne ARCserve/Open
needs to offer a non-proprietary tape format and could stand better data
migration tools as well as a boost in performance. In addition, we found
Dallastone DallasTools and Workstation Solutions Quick Restore lacking in
performance.
Legato Systems Legato NetWorker
 
NetWorker has excellent functionality. Strengths such as its parallelism
feature, real-time monitoring and some very flexible configuration options,
combined wi
th its multiserver and backup formats make this product unbeatable.
Although NetWorker's performance was second to PDC, it would have clearly
out performed PDC had our testing methods included the
parallelism feature.
NetWorker's parallelism feature provides support for a maximum of 32 client
backup/restore sessions to 16 different tape devices. This is achieved by
interleaving sessions from each client. Interleaving decreases the overall
backup system's idle time and, therefore, drives the tape library devices
at their peak. NetWorker achieved an average of 430 KB per second transfers
to a single drive without parallelism and at times even exceeded 500 KB
per second (the EXB8505 tape drive operates at 500 KB per second without
compression). With parallelism on, we had peaks that approached our Ethernet
bandwidth limit! NetWorker was just plain fast. We speculated that we could
easily achieve 2 MB per second backup speeds with our four-drive EXB-440
if not for our 750 KB per second bandwidth restriction.
NetWorker, like Spectra Logic, also receives high marks, since it can back
up in a variety of formats: cpio, dump and tar are all supported. That's
good news for administrators
who don't want to lock themselves into a proprietary
backup format. NetWorker also provides a multiserver configuration that
allows you to switch to a completely different media server on your network,
schedule backups and perform restores as well as control the media server.
Although there isn't a command line equivalent for every GUI menu item,
NetWorker has a great ASCII equivalent of its menu system that allows you
to set almost every parameter and function that you'd find in the GUI interface.
This is a boon to those administrators who want to handle backups or restores
remotely via Telnet. NetWorker supports all the big Unix versions, NetWare,
Macintosh and Microsoft Windows95 3.11 and NT.
PDC BudTool

PDC demonstrates the strength of Unix really by incorporating standard Unix
backup tools like dump and restore into BudTool. Considering BudTool's use
of these to
ols, it had amazing performance. Another advantage of BudTool
is that it does not require client-side software. It utilizes Unix rdump
and rmt, so almost everything is already installed.
BudTool's only drawback was a few sloppy feature implementations. But these
did not seem to have any implications for performance or useability. We
found that PDC spit out errors from its restore process, due to the fact
that terminal I/O from the process was not captured and suppressed from
the user's view. PDC advised that restores would still work despite this
error and that it would be addressed in a future release.
BudTool is a very flexible and accommodating package that enables backups
and restores on any Unix flavor or variant, due to its ability to use standard
tools. However, with respect to server platform and backup hardware, you're
limited to Unix vendors that PDC supports directly: SunOS, Solaris and SGI
to name a few.
PDC presents a very simplistic approach, dividing setup and d
etailed operations
into two separate GUIs called btadmin and xbt. The btadmin utility is used
to initially configure our Exabyte library, backup strategy and clients
to back up. Backup and restore operations are performed via the xbt utility.
A major advantage of BudTool is that it doesn't require client-side software.
But, the client installation did require us to put an .rhosts file on each
client to allow root or backup-capable access from our backup host. For
those who are more security conscious and not comfortable using .rhosts
files, BudTool probably should not be your first choice. But from a management
perspective, having no client-side software means that you never have to
update it when the underlying OS changes.
Scheduling automatic backups with BudTool was straightforward, due in part
to its simple point-and-click interfac
e. It allows the file system to be
backed up to specific drives or libraries and supports parallel backups
to individual drives in the EXB-440. BudTool,
like the rest, also had a
tool that reset the database of tape locations within the EXB-440, in case
library tapes were changed or moved outside BudTool operations. We found
that BudTool warned us via e-mail and did so more reliably than the rest.
Restores were very simple. BudTool has an online database of all the hosts,
directories and files it has backed up and provides a point-and-click hierarchical
interface to manage them, similar to what ARCserve and Quick Restore provide.
The Solaris 2.x version of the restore portion of BudTool required us to
create a /etc/restore link to /usr/sbin/ufsrestore (the SunOS 5 version
of restore now) for it to work. Once linked, the restore worked. However,
we discovered misleading /dev/tty errors that may cause the inexperienced
user to think that the backup did not occur. Some of these error messages
are documented, but we thought this could be programmed better.
System administrators will be happy to note that BudTool can be configured,
like NetWo
rker, Alexandria and ARCserve/Open, to allow any user to browse
the restore database and select files for retrieval. That's great for those
times when you have more pressing work.
Another strong point of BudTool is its cross-platform restore capability.
As long as dump formats between Unix are compatible, it is easy to dump
and restore files between various Unix flavors--an excellent function if
you're migrating between Unix brands. However, this is not all encompassing--HP-UX
and AIX cannot be restored to SunOS machines.
Spectra Logic Alexandria Backup Librarian
Spectra Logic provides a seriously powerful package that would have done
better if it had been easier to use, performed better and supported more
than just Unix. We like the power it offers, but we found it very complex
and tedious to configure. We knew testing Alexandria would
be challenging
when we opened its two-inch thick, three-ring documentation notebook!
Alexandria provided great hardware support, g
ood security and fault tolerance,
along with a very detailed operating interface. If you need good security
and data redundancy, are willing to take the time to learn Alexandria's
complexity and have only Unix clients, this may be the package for you.
Alexandria comes on strong in its hardware support through its use of a
unique Serial SCSI Command Language (SSCL) interface. This interface, which
connects to the SCSI bus, allows Alexandria to support more than 30 different
automated media libraries. The SSCL interface addresses libraries through
the server's RS-232 interface, avoiding the need for special device drivers
or kernel modifications.
Alexandria installation was involved and required planning. We found we
needed to create a special user and group on our backup server, as well
as on each of our clients. We were not comfortable with the fact that Alexandria
required us to find a matching UID and GID on both server and client systems
before we could install it safely. If the insta
ll is started without this
pairing, you could be in for quite a headache trying to find a free UID
and GID later.
Configuring the tape library through the SSCL was fairly easy. However,
we did run into problems when we inadvertently placed a space in the name
of our EXB-440. Alexandria broke the string into two and used only the first
portion of the name. In doing so, it failed to acknowledge the EXB-440.
After renaming it, without the space, we were able to continue configuring.
Alexandria, although filled with many menus and options, allows every menu
option to be executed from the command line. This provided great control
when accessing any backup machine remotely and allows custom scripts to
be designed and used.
Security is strongest in Alexandria. Other packages don't give as much attention
to security levels and access. Alexandria pays close
attention to regular
Unix permissions and improves security by allowing the backup administrator
to grant or deny access to backu
p functions based on Unix UID and GID. As
many as 13 different types of access can be granted or denied to all users,
groups or a specific user to provide very flexible or exact types of access.
We found Alexandria extremely tolerant of drive failures. Between backups,
we removed one of the defined tape drives and Alexandria successfully labeled
this drive unavailable, took the device offline and continued our backup
operation on a secondary drive. For sites that require multiple media servers,
Alexandria fits right in.
Cheyenne Software ARCserve/Open
Cheyenne's ARCserve/Open wasn't perfect, but its simplicity gave it merit.
It was easy to install, and with it, easy to back up and restore data. In
fact, these three steps were done within the first hour of opening the box.
Poor performance numbers and its proprietary format were its downfall. ARCserve
is the right product if you're looking for a simple and easy-to-use package
and want to stay as far away from Unix as possible w
hile still managing
Unix data.
Cheyenne's motto seemed to be "keep it simple." We liked its simple
setup and its ability to be easily configured. Once the software was installed
and started, it seemed like we left Unix behind and we were back on a PC
or a Mac. In fact, the interface is almost exactly what you'd find in the
PC Admin tool for ARCserve for NetWare.
Configuring the server to recognize clients did require some data entry,
which could be painful if you have a large number of hosts or clients to
back up--all six of the packages evaluated required this to some degree.
Once a host is entered into the remote client database, it was then immediately
available.
Configuring the EXB-440 was relatively easy, but took a long time when it
came to media formatting. One thing that bothered us was that there was
no notice announcing, "This is going
to take hours to label all 40
of these tapes!" This might cause inexperienced administrators to think
the syste
m is hung unless they actually looked at the library. Additionally,
ARCserve fell short in this version by not providing bar code support for
the EXB-440. However, Cheyenne does support bar codes on other Exabyte model
libraries.
Starting a backup or a restore was so simple that it is almost not necessary
to describe it. ARCserve allows you to browse the entire file system of
its defined remote clients and click on whatever partition, directory or
file you'd like to select. Restores were equally simple. Every file, directory
and partition was viewable within a restore tree. Again, by simply clicking
on what you want restored from a particular tape selects it for action.
Like Workstation Solutions Quick Restore, PDC BudTool and Alexandria the
destination of the restored files is fully under your control.
Dallastone DallasTools
Dallastone offers a powerful, yet maturing product. It provides some features
that set it apart from the rest. However, we found performance less than
av
erage and its interface is not as friendly as the rest. DallasTools most
powerful features are its command line capabilities, its restore independence
and multiserver configurability. In the end, Dallastone did poorly because
of slow performance and a lack of good reporting and progress monitoring
tools.
DallasTools ships with a rather clumsy GUI that is not much more than a
front end to its command line tools. DallasTools did facilitate configuring
and managing the EXB-440 fairly well, but it isn't very intuitive when it
comes to backing up and restoring files. It refers to the EXB-440 as a "stacking
device" and refers to the archive restore as the "librarian."
We felt these terms were misplaced. Isn't the "librarian" the
EXB-440? Isn't the archive restore where you restore files?
We had a few problems installing DallasTools and got lost when looking th
rough
its documentation for help. Nowhere in the table of contents could we find
anything that referred t
o backing up a client. There were sections on the
write functions of DallasTools, but we had to read the documentation in
detail to understand how to get a file written to tape. We found ourselves
spoiled by the point-and-click approach used by the other packages, allowing
us to bypass detailed reading of documentation.
Defining backups with DallasTools is a real convenience. Using the main
GUI window, we could enter any combination of host files or directories
to back up. It made grouping clients or similar files among clients easy.
Once our desired data was selected, the backup criteria was set. This section
was separate from the scheduler, where we thought it made more sense to
be located. We did finally get a backup going, but performance was very
poor.
After allowing Dallastone access to our test environment, it raised performance
to an average level, but was unable to get any more performance out of the
configuration. These second performance numbers are reported in our performance
chart, below. Dallastone was quite surprised by our performance numbers,
stating that it was far below what both it and its customers expect. Our
testing methods were strict and looked at performance for an entire host.
This forced Dallastone to encounter different types of files, both small
and large. Dallastone performed very well with large files, but when performance
was measured for an entire host, its performance numbers plummeted.
Dallastone had little reporting and monitoring access. You could monitor
the activity of the backup using the Unix tail command with the full logfile
(which generated a rather unwieldy 29-MB file when we tested our four clients).
You could also have this file mailed to an operator. To determine performance,
a bit of Unix scripting was needed to generate a report. Dallastone provided
no built-in functions to depict performance, and therefore we found ourselves
writing
a short Bourne shell script to determine performance.
Similar to NetWorker and Alexand
ria, Dallastone has a good thing going with
its server/subserver (multiserver) technology. For environments that have
a number of subnets to back up, DallasTool's subserver strategy is a good
one. This strategy allows for the set up and administration of all DallasTools
servers (called subservers) to be carried out on a single controlling server.
With this technology, tape devices attached to subservers can be controlled
by a central server. Each of these subservers can have an independent client
set.
Workstation Solutions Quick Restore v2
Workstation Solutions' Quick Restore provides a solid product, but it has
some problems. It is user-friendly, has good restore independence and was
the easiest of all to install a client. However, Quick Restore had average
performance and developed problems on our multidomain/multisubnet network
during configuration. We were forced to make changes in our DNS to accommodate
the shortcomings in Quick Restore.
Quick Restore has an easy configur
ation and operation GUI. The setup of
the server and the EXB-440 went smoothly. However, during our install of
the clients, only three out of four clients installed. For the three that
did work, we simply opened the client machine for root access via .rhosts
(for the security conscious, there is another way) and ran the installer
from the server. It performed the installations and placed all required
files on the client. Our fourth client confused the install program in a
way that was undetermined. Even with Workstation Solutions' help, we could
not duplicate the first three simple network installs. After numerous attempts,
we resorted to FTPing the files to the client for installation.
Once our four clients were installed, we then developed problems with two
of them. We isolated the problem to Quick Restore's inability to cope with
multidomain or multiple subnetworks. Our server was in a separate subnet
than th
e clients for our testing and we had multiple domains involved as
well. Our server and
two of the clients were in domain_a.com while the two
other clients were in domain_b.com.
As we tested, we could not initiate a backup of our SPARC Classic and SPARC
20 clients from the server. However, we could initiate a backup from the
clients to the server. Even after adding A records to our DNS so that all
hosts were in a single domain (we did not change the actual client hostnames),
we could not back up all the clients from the server. Quick Restore's performance
numbers may have been higher if we had been able to initiate all the backup
trials from the server. Workstation Solutions reports this problem will
be addressed in its next release. Otherwise, Quick Restore had good restore
independence, since it stored data in POSIX-compliant format, like Alexandria,
NetWorker and BudTools can.
Quick Restore stands strong when it comes to ease of restore. It has a great
browser for picking files from the backup archives. It actually has two
modes to restore files: a basic mode and a power
mode. Basic mode is a simple
"pick a file" and restore it, while the power mode is one that
enables the building of a list of clients/paths to restore. It then restores
them all in one step. That's very convenient if you're doing a large number
of restorations.
Aaron C. Young (ayoung@nysernet.com) is a systems administrator at NYSERNet.
With 10 years of experience as an Internet provider, NYSERNet specializes
in Web hosting, access and customized Internet software solutions.
Getting Backups Against The Wall: How We Tested
To test these backup products, we used five Sun Microsystems machines--four
with varying amounts of data to back up and one acting as a backup server.
These machines each held between 300 MB and 3,000 MB of data.
Our dedicated backup server, a SPARC 20, ran SunOS 5.5. It also hosted an
Exabyte EXB-440 8mm tape library containing four 8505XL drives. Ou
r clients
included a SPARC Classic which ran SunOS 4.1.3_u1, two SPARC 4
s, both of
which ran SunOS 5.5 and a second SPARC 20 that ran SunOS 4.1.3_u1.
Our initial goal was to run four backups in parallel, one to each of the
drives in the EXB-440. But our available bandwidth was a 10 Mbps Ethernet,
with a 1,250 Kbps theoretical throughput. Additionally, with this Ethernet,
we estimated that a practical upper limit for data transfers, considering
overhead and collisions, is about 750 Kbps.
Given this, and the speed of our test machines, it became clear that four
parallel backups would easily max-out our 750 Kbps available bandwidth.
Each tape drive within the EXB-440 can achieve a sustained rate of 500
Kbps. If each software solution achieved even half of this rate, it would
easily exceed the 750 Kbps approximate limit of our network.
So, we chose to test using sequential backups, each performed to the same
EXB8505 tape drive. Although we would have loved to exercise the EXB-440,
this environment allowed us to identify how each of the various backup
packages
performed with the unit.
May 1, 1996
|