|Next Page||Prev Page||Main Index||Home|
Like most database management systems, MPbase's characteristics
depend to a large degree on the specific data and access characteristics.
However, unlike most database management systems, there is no
need to make tradeoffs. Each characteristic can be determined
individually and independently. In addition, multiple different
options can be selected for a single characteristic. For example,
a single copy of a single database can be accessed by one method
tuned for On-line Transaction Processing (OLTP), and at the same
time by a different method tuned for batch list processing.
1) Q I have an OLTP system supporting 25,000 simultaneous users
with a sub 2-second response time. What will an MPbase
do under similar circumstances?
A If there is no need to improve the above numbers, you could
expect the same performance for between 1/10 and 1/100 the current
cost. At the same time, you would gain the ability to reduce response
time and/or increase the maximum number of users to any level
at any time. Given the needed hardware, 250,000 users with sub-second
response time would be a reasonable expectation.
2) Q My DASD farm has 24 terabytes of data and costs $24,000,000.00/year
in lease, maintenance and environmental expenses. What about yours?
A The MPbase stores the data in a highly compressed format.
Therefore, assuming your data is in a flat ASCII (text) format,
you should expect to need about 300 gigabytes of workstation disk
per copy. At 1996 prices, this would be about $35,000.00/year
per copy which includes the supporting I/O cluster. Also, this
type of disk has no special environmental or power needs. This
produces significant savings over the traditional machine room.
Note, that at this price, multiple copies are recommended.
3) Q My data store needs 24 terabytes and grows at 15% per year.
How does an MPbase scale at this size and rate?
A MPbase scales by adding "nodes." A node is
defined as one processor and a set amount of disk capacity. The
ratio of disk/processor is determined by the access requirements.
Assuming for a moment that the above system starts with 100 nodes
you would need to add 15 nodes per year. There is no upper limit
to the number of nodes that can be added to one system.
4) Q I have 24 terabytes of storage. It takes 12 weeks for a full
backup and two hours per day for an incremental backup. How would
your approach perform under similar circumstances?
A If there is no need to improve the above numbers you would expect
the same performance for less cost. However, with the assumptions
from the answers to questions 2 and 3 above, if required, a full
tape backup could be made in as little as 30 minutes. This would
require one tape drive per node (initially 100). In addition,
a second disk copy could be kept in sync in real time.
5) Q My system runs 24 hours/day and seven days/week. How would
your solution protect against outage?
A An MPbase with two or more copies of the data is fault-tolerant
at the node level. This allows configurations that have no
single point of failure. This means any node may be powered
down at any time and still have full access to the data. In addition,
large clusters normally have a number of "spare" nodes.
These spares can automatically assume the role of any node that
may fail. Recovery involves rebuilding the data files from the
second online copy. Once this is complete, the cluster is back
at full strength, with one fewer spare, naturally.
6) Q What is the incremental cost of a complete disaster recovery
site and operation?
A There are two main modes of disaster recovery available to MPbase installations. In addition, any combination of these two will also work.
7) Full hot site with the files in sync, real time.
With this option cut-over time measured is in seconds.
The cost can vary from the cost of the connecting data links,
to the link costs plus the full cost of the cluster. This is based
on how well the system can tolerate less-than-full performance.
With three working sites, loss of any one site reduces the total
capacity by only 30%. If the cluster is designed with 30% extra
capacity, the net effect to the users is zero.
8) Store the small set of components that cannot quickly be rented
and rent the rest when needed.
With this option cut-over time is one or two weeks. This depends
in large part on the backup/restore option selected.
Cost could be as little as $20,000.00 until needed. The cost to
rent the "commodity class" hardware for a full year
should about equal the purchase cost.
9) Q I have fifty programmers, systems people, and operators who
report to me and make up my domain. What would your approach do
to help me significantly reduce costs and still maintain my department's
significant contribution to the organization?
A First, MPbase can substantiality reduce your costs. This
reduction would come from hardware, system software, user software,
maintenance, environmental, and operational costs. At the same
time, significant new functionality and capacity could be added.
Second, as staff is freed from day-to-day operational tasks, they could be re-deployed to increase the value of your department within the company. In the current business environment with an emphasis on re-engineering, operational units are valued by output not input.
|Next Page||Prev Page||Main Index||Home||© 1998-2004 NPSI|