Home Search
IT
Resources |
Home |
Contact Search
|
| IT
Resources |
||
|
Appendix 2 - Server Infrastructure Options
The second approach is to use a distributed file server
model, utilising a number of mid range servers, each capable of file
serving approximately 100-250 clients. Such machines could be deployed
on an 'as needed' basis, but remain centrally supported and maintained
by Infrastructure Systems. There is a range of possible iterations of the second
model, which offers flexibility in terms of deployment and reduces the
up-front cost faced by the University in investing in this system. In
addition in the event of failure of a single small server only a small
number of clients would be directly affected. However the disadvantage
is that this model is more complex from the systems perspective. In
addition, individual units would not be as reliable as a smaller number
of large servers although this could be offset by greater redundancy
and by carrying spares. Backup capacity and assumptions
Currently Retrospect Remote backups range from 5 MB
on some client machines to 2 GB on others. Typically the amount is less
than 100 MB, and it is reasonable to assume that this would be the average.
On a server there is less likely to be duplication and unnecessary backing
up (eg of applications) so we can assume that the space required will
be lower, thus leading to a saving of space in backup capacity. For the purposes of this proposal we will assume that
an average allocation of 100 MB per staff user is adequate - it is a
figure commonly accepted within industry at this time. For reliability
servers will be assumed to employ RAID Level 1 (striping/mirroring)
which halves the storage capacity on the server (eg 100MB data requires
200MB server space). The operating systems and NDS overhead will be
in addition to client data space. All servers would be recommended on
the basis of current capacity and future expandability to ensure maximum
return on investment. For the student desktop services implementation in
approximately 2 years' time it is assumed on current capacity that 20MB
per user is adequate, or 10MB with mirroring and striping. This is an
operational decision that can be made closer to the time based on the
first two years' experiences. Capacity is calculated on an expected
capacity of 15,000 students (not FTE) actually using the system. Honours
students may require an additional allocation but they are a small proportion
of the overall student population. It would be advisable to budget for future expansion in disk capacity every two years, and a replacement/upgrade cycle of 3-4 years for the server hardware. By staging the student services implementation approximately 2 years after the commencement of the project it would be economical to allocate the initial hardware purchase to student services then, and purchase newer, faster equipment for the existing staff infrastructure. Thus each hardware cycle could be stretched to 4 or 5 years for the NDS infrastructure by upgrading 50% of equipment every 2-3 years, attaining the best possible performance for the backbone NDS system at the same time as getting maximum value from the hardware investment. Option 1 - Large capacity high availability file server:
File serving using SAMBA on a Unix host is fairly light
load work. We can specify one large capacity machine (eg: Sun Enterprise
450) in Hobart and one in Launceston which would each be capable of
carrying at least 1000 client connections. This option offers several advantages. Topographically,
NDS only requires one server in each site, which we would provide through
one machine in each of Hobart, Launceston and Burnie. This has the advantage
of presenting Systems with fewer machines to maintain, thus requiring
less man-hours, and greater individual reliability. These machines would
be high availability machines using hot swappable power supplies and
hard disks, meaning that any form of catastrophic hardware failure is
as unlikely as possible. Further advantages are realised in cost per client.
With a massive capacity for file storage and serving such machines would
work out as being cheaper per client to purchase and maintain than a
larger number of smaller machines, as well as offering greater longevity.
These advantages are not however immediately realisable as they require
an economy of scale. Typically a client base of 1000+ clients would
be needed for each machine before cost benefits would be evident in
comparison to smaller servers. Disadvantages include the high capital cost of each machine, and the risk, however small, attached to basing a lot of data in one location. If there was a catastrophic server failure it would affect many areas simultaneously. Option 2 - Distributed medium capacity file servers
Typically a medium capacity server could be an Intel
machine (ie standard PC) running one variant or another of Unix- either
Solaris x86, Linux or FreeBSD.
It would be possible to do this task running Windows NT Server
or Novell Netware however Unix offers significantly greater performance
on the same power of machine and in addition will integrate better with
existing equipment run by Systems. Other suitable hardware platforms would be either small capacity Sun machines or even other systems such as those offered by Cobalt (see attachment D). At this stage however it is most likely that the solution favoured would be Intel or Sun. 2a - Sun platform
A medium capacity Sun file server would be expected
to handle easily 250 clients, or 25% of the capacity of the high end
option outlined above. For details, see attachment B. The advantages
of this approach would be that with a lower capital cost per server
it would be simpler and cheaper to add additional capacity as required,
and to upgrade in the future. It would lower the initial capital cost
of the project significantly. Other advantages include greater flexibility
and greater redundancy - if one server goes down then there are others
to carry the load. Disadvantages of this approach however include a probable
higher overall cost per client as more machines would eventually be
required and would therefore probably exceed the total cost of the larger
machines for the same file serving capacity. Furthermore, more machines
take more maintenance from the Systems group and therefore will impose
a greater overhead on Systems personnel. Individual machines may not
be as reliable as the big servers as they do not have built in second
power supplies, or hot swappable drives. For serious high availability
the big machines are undoubtedly the best. 2b - Intel platform
High end Intel powered machines are typically cheap
to purchase and will provide good performance and reliability. They
are in addition cheaper to upgrade and maintain however the available
operating systems are their limitation. Windows NT Server does not really
offer the performance of its competitors and is therefore not recommended
for this purpose. Novell Netware is more efficient and will perform
well, however Netware experience and knowhow within ITS is limited,
and training is expensive and specialised. Solaris is the preferred
platform as the Systems group is well equipped with Solaris expertise.
Solaris is available for the Intel platform however it does not currently
cater for dual processor machines which is a serious limitation. In
addition it is a new product and does not have the maturity of Solaris
on the Sun platform. In testing so far however it appears quite manageable
and further testing may yet show this to be a worthwhile product. The
final choice is Linux, a freeware Unix variant that has made great progress
in the last two years. Linux will cater for dual processor Intel boxes,
and is well respected and supported in the Unix community. It is probably
too soon for this product to be offered as a suitable option. As mentioned above, the initial costs of deploying
Intel machines would be significantly lower than a big Solaris machine,
however as the load increases this differential may well decrease. Advantages
include easy servicing/upgrading and low capital costs. Disadvantages
include lower reliability of hardware in comparison to high-end Sun
machines, lack of experience with Intel machines within the Systems
group, and lack of expertise in the various operating system options
available. 2c - Cobalt
Cobalt is a small company that builds purpose made small capacity network servers running Linux on a MIPS processor. This is an unusual hardware configuration however the products are well designed and are worthy of further consideration. Advantages include very low capital cost per unit, easy deployment, high redundancy (ie not all eggs in one basket) and easy management using purpose made tools. Disadvantages include little known hardware platform, small size of supplier and the need to have many small units distributed according to need. It is not at this stage confirmed whether Linux will operate with NDS although it is believed to be the case. Recommendations:
The preferred solution would be the following implementation,
which is a combination of the medium and large server capacity options.
Paired Enterprise 250s are preferred to a single Enterprise 450 as they
offer greater flexibility and future clustering options. Implementation could be staged in line with NDS implementation
if NDS is implemented over a period of time, or installed immediately
to provide the framework both for NDS and a client/server implementation
of the lease scheme. These are the basic requirements that would fulfil
our needs for a full implementation of NDS across the university, and
cater for implementation and expansion of the Client/Server lease scheme.
All these options have expansion capability to allow for future growth
in demand. Hobart main campus: Launceston main campus: Burnie main campus:
|
