Server Architectural Models (Draft)
A number of design decisions must be made when developing or enhancing information services.
Some of the factors to consider are:
- Availability (or uptime)
- Manageability
- Performance
- Scalability
The importance of each factor will depend on the user's needs and how the IT organization can
meet these needs within the context of the organization's technical and operational architecture.
What is Availability?
Availability is the time during which a device, such as a computer, or a service, such as web server, is functioning or available for use.
Availability depends on many factors including software stability, hardware load, and infrastructure reliability.
As such, there is no "industry standard" for availability. The table below illustrates some common availability metrics.
For example, a "five-9s" metric is available 99.999% of the time which is to say that the system or service is down
for just 5.25 minutes in a year. As a comparison, a 99.000% rating may be down for a total of 87.6 hours in a year which
is an average of about 1 hour and 40 minutes per week.
| Percent Uptime | Minutes Downtime* | Hours Downtime* | Days Downtime* |
| 99.999% | 5.25 | < 1 | < 1 |
| 99.990% | 52.56 | < 1 | < 1 |
| 99.900% | 525.6 | 8.76 | < 1 |
| 99.000% | 5256 | 87.6 | 3.65 |
| 95.000% | 26280 | 438 | 18.25 |
| 90.000% | 52560 | 876 | 36.5 |
| 85.000% | 78840 | 1314 | 54.75 |
| 80.000% | 105120 | 1752 | 73 |
| 75.000% | 131400 | 2190 | 91.25 |
*(all figures assuming 365 days per year, 24 hours per day, 60 minutes per hour)
We have defined seven server architectural models ranging from very high availability servers
to low reliability servers. The seven models are described below in decreasing order of reliability.
Model 7
Model 7 is the highest availability server architecture with the goal of achieving 99.999% availability.
This model has no single point of failure.
Redundant servers are clustered together with either automatic load-balancing capability or automatic
fail-over capability. The servers have redundant connections to different switches within the
IP network, have redundant connections to the high-availability Storage Area Network (SAN),
use a standard hardware and software platform,
and are housed in an environmentally-controlled computer room.
All devices from the core routers to the disk drives
have appropriate hardware and software maintenance and
are powered by Uninterruptible Power Supply (UPS) equipment.
The UPS equipment is either connected to an emergency generator or has a battery time of 4 hours.
All components participate in a comprehensive enterprise-class management and monitoring system
and all components participate in a periodic hardware and software refresh program.
The server and storage system undergoes periodic testing for automatic failover assurance.
Backups can be disk to disk then to tape for very fast backups.
Model 6
Model 6 is a high availability server architecture with the objective of obtaining 99.9% availability.
It is very similar to model 7 except that
there is a single point of failure in the design, such as the servers may connect to a single LAN switch in the IP network.
Backups can be disk to disk then to tape for very fast backups.
Model 5
Model 5 high availability server architecture with the goal of obtaining 99.9% availability.
It is very similar to model 6 except that
the storage is local RAID instead of the high-performance SAN storage.
Backups are traditional disk to tape over the network.
Model 4
Model 4 has the goal of achieving 99.0% uptime.
It has multiple single points of failure and relies on
a single server. The advantage of model 4 is the high-performance SAN storage system.
Since storage is well-protected it is possible to have a backup server available
that can replace a failed production server but this cannot be done automatically.
Backups can be disk to disk then to tape for very fast backups.
Model 3
Model 3 has the goal of achieving 99.0% uptime.
It is a traditional single server model. The data is protected by RAID technology and
thus the loss of a single disk drive does not take the system down. The
server is housed in an environmentally-controlled computer room with proper UPS, HVAC, etc.
The server is refreshed on a periodic cycle and has a common hardware and software platform.
However, the system has a single network connection and a single server.
A backup server may be available but has to be manually placed into production.
Backups will occur over the network and will likely be disk to tape.
Model 2
Model 2 will likely achieve 99.0% uptime but there is no guarantee.
Model 2 is similar to model 3 except for the disk storage. In this model the disk
storage is not protected by RAID and thus loss of a disk drive requires manual intervention
in order to restore data to the system. The data restore is from a tape which is inherently slow.
Model 2 retains the advantage of a standard hardware and software platform.
Backups will occur over the network and will likely be disk to tape.
Model 1
Model 1 will likely achieve 99.0% uptime but there is no guarantee.
Backups will occur over the network and will likely be disk to tape.
|