The Infrastructure
Upgrade: The Story Continues
By
Annie Yu
|
|
|
In
the previous issue of Network Computing, we have touched on a
few aspects of the large-scale plan for IT Infrastructure Upgrade,
namely what actions should be taken to remedy the weak points
of the Campus Network, the addition of resources to some of the
critical central servers and the largest ever PC upgrade exercise
which was conducted in full swing only recently. In order to complete
the picture of the plan, we shall examine in turn each of the
other areas that also require immediate attention. They are:
1) Upgrade
the network infrastructure and bandwidth
Over the last
15 years, the CTNET, CityUs campus network, had undergone
several phases of development in order to cope with the increasing
needs and to adopt new technologies. Currently, the network consists
of two parallel backbones, one for carrying ordinary data and
the other for video. By adopting this model, we can be sure that
optimal performance can be achieved when running services like
CityVoD, video conferencing, video and TV broadcast, digital image
library and multimedia applications without affecting the network
traffic of services provided on the data network.
At the moment,
the campus network is still sufficient to handle the daily traffic,
but what about in a few years time? We can foresee that
a humongous amount of bandwidth will be required, which might
result from an increasing number of network nodes, increasing
processing power of PCs and workstations, rapid proliferation
of multimedia applications and so forth. A major upgrade is considered
necessary and its benefits are manifold as described below:
a.
The existing network is found to have problems or limitations.
To list a few:
-
The
2 backbones each plays a specific role. Mutual backup is therefore
extremely difficult if one of them fails.
-
Local
Area Network Emulation (LANE) services impose a significant
overhead to the ATM core, which render the ATM backbone unable
to support high volume of broadcast and multicast traffic.
-
One-armed routers are used to forward intra-Virtual Local Area
Network (VLAN) data traffic on the ATM network. These routers
are relatively slow compared with layer-3 switches and therefore
have become a bottleneck for delay-sensitive and bandwidth hungry
applications.
-
Quality of Service (QoS) is not supported by the existing ATM
network, which is based on LANE 1.0. It must be upgraded to
MPOA, which includes LANE 2.0, in order to provide QoS.
-
The backplanes of the ATM switches are limited to 3.2 Gbps,
which means future upgrade of the ATM network to support gigabit
networking will be a concern.
-
An average Ethernet or Fast Ethernet port on an ATM switch is
comparatively more expensive than a 10/100 Ethernet port on
a layer-2 or layer-3 Ethernet switch. Likewise, its annual hardware
maintenance cost is higher.
-
ATM technology is superseded by Gigabit/Fast Ethernet technology
in the majority of new campus network installations/upgrades.
Putting additional investments on a technology near its end
of life cycle seems inappropriate.
-
Existing installed multi-mode fibers are not capable of supporting
gigabit speed at a distance of over 300 metres.
-
Network monitoring tools are inadequate for the collection,
comparison, and analysis of network performance and operational
metrics. Shortage of manpower for network monitoring is yet
another issue.
It is envisaged
that most of the aforementioned problems can be resolved by the
upgrade.
b.
The upgrade will put the University into a competitive position
against other local universities, most of which are in the process
of upgrading their campus networks (Specifically, HKUST will provide
100Mbps for each network connection, and 1000Mbps for each network
uplink. It sounds overkill, but the provision of network bandwidth
is a factor being considered when a comparison is drawn between
the universities).
c.
The 2 existing backbones will be merged into one based on Gigabit/Fast
Ethernet and layer-3 switching technologies, resulting in a simplified
network architecture which will ease network management and support
efforts, as well as reduce TOS (total cost of ownership).
d. QoS
will be supported, and therefore delay-sensitive applications
such as Video-conferencing VoD, TV broadcasts and Voice-over-IP
can be better supported by the network.
e.
Gigabit Ethernet uplinks will be provided between wiring closets
and the core switches. The fiber cable plant will be upgraded
to embrace single-mode and the new 50 mm multimode fibers. The
new cable plant will be able to support gigabit networking as
well as the forthcoming 10 Gbps Ethernet standard.
f.
Network monitoring tools will be used to monitor the network and
collect real-time or historical metrics. Such metrics can be used
to establish baseline measures and watch for trends.
g.
Additional VLANs can be supported. Each department may have its
own VLAN(s) if necessary.
2)
Consolidation of Departmental Local Area Network (LAN) Servers
At the moment,
the existing 55 staff departmental LANs consist of 60 NT servers
and around 3000 Windows 98 clients. Among the 60 NT servers, 5
have been configured as a single master domain with one as Primary
Domain Controller and 4 Backup Domain Controllers. The master
domain servers handle user account maintenance, logon validation,
WINS, central software backup services for Windows 95 and 98,
central Intranet menu, and as backup resource domain servers for
departments. One disadvantage of the latter setup is that disk
space on NT servers are more difficult to expand and less reliable,
and users therefore tend to store their data on other central
servers such as email servers. Furthermore, huge disk size and
disk space for data and programs are increasing so fast that existing
tape drive will soon no longer be able to handle their backup
and restore within reasonable time.
The remaining
55 NT departmental servers, that serve around 60 departments,
have been configured as resource domains which support departmental
software installation, file and print services, departmental Intranet
menu and user-specific file sharing.
Depending
on the licenses, usages, and sources of funding, some client software
with network version are installed on departmental servers while
some with individual licenses are on clients PCs. Network
version of software owned by one department will be installed
only on the departmental server of that department. Thus, some
popular software inevitably will have to be duplicated in each
of the departmental servers, thereby increasing the supporting
effort.
In order to
eliminate the shortcomings of the existing setup and enhance our
LAN services, it is proposed that:
-
The master domain servers are to be replaced with servers running
Clustering and NSS technologies. With these technologies, chances
of server down or disk failure will be significantly reduced.
Expansion or dynamical allocation of disk space will be much
easier and more flexible. Furthermore, additional critical functions
of departmental servers can be replicated in these servers for
fail-over purposes.
-
All popular software be relocated from departmental servers
to the Clustered servers such that only a single copy for each
software needs to be maintained thereby reducing the supporting
effort.
3) Establishment of a Central Network Storage System (NSS)
Today, servers
across the LANs or Wide Area Networks (WANs) need to provide fast,
reliable, ubiquitous access to data as required by various applications.
Moreover, our mission-critical applications, in particular, require
a data storage that can be operational all year round and can
be expanded easily without interrupting any application currently
running. At present, all our data is stored in the direct attached
storage. Centralised storage systems such as the network-attached
storage (NAS), the Storage Area Network (SAN) and Enterprise Storage
Network (ESN) are now available in the market. Adoption of a Central
Network Storage System (NSS) is definitely the direction to go
and its benefits are quite apparent:
a)
High Availability and Disaster Recovery
NSS has the
ability to make any-to-any connections among multiple servers
and storage devices. They can create a shared pool of storage,
which can be accessed by multiple servers through multiple paths,
resulting in high availability. NSS therefore provides an excellent
environment for clustering that can extend to dozens of servers
and storage devices. If storage devices fail, servers continue
to function because another copy of data is still available through
NSS.
Through NSS,
many vendors now offer 100% data availability; this makes long
backup time less critical and disaster recovery less an issue.
b)
Backup
NSS simplifies
the backup by sharing tape drives among servers. It also allows
a single backup be handled by multiple tape drives simultaneously,
thereby reducing the backup time.
c)
Data Sharing
NSS allows
multiple distributed servers to concurrently access a centralised
storage for data sharing applications. With NSS it is easy to
deploy more idle servers to handle those mission critical applications
with a short peak usage periods (e.g. students registration in
summer time) and to resume the server number to normal once the
peak usage periods are over.
d)
Management and Future Expansion
Storage devices
could be distributed through a network yet managed from a central
point through a single management tool. This will lower the cost
of storage management, standardise the control of the administrator
and hence provide better reliability and availability. New devices
can also be added online without disrupting data access.
It is planned
to implement the NSS first on all mission-critical central hosts
such as web servers, email servers and servers of the departmental
staff LAN, and later to cover most of the central requirement
for storage.
|