Network: Network is most important for successful implementation of data replication solution. Network requirements and constraints define appropriate data replication solution to be used. Network requirements can be split into latency and bandwidth requirements. Combination of latency and bandwidth define capacity of network.
Every application do not have same network requirements. Some require lower latency and other require high throughput. A network can be of any types below
- High bandwidth with low latency
- High bandwidth with high latency
- Low bandwidth with low latency
- Low bandwidth with high latency
I found a nice article that differentiates between Latency and Throughput.
http://dsoguy.blogspot.com/2007/06/latency-v-throughput.html
Applications that need low latency connections:
- Typical Internet browsing would typically need a low latency connection as time to load page should be as fast a possible.
- Online games and typically multi player games would again need a low latency connection even though bandwidth requirements may be low.
- Any monitoring software(s) would also need a low latency connection. Such applications monitor state of service / software and update flags or status values at central repository. So, data volume would be less but for proper up to date real time monitoring latency should be as low as possible.
Applications that need high bandwidth connections:
- FTP connections need high bandwidth but even less latency would work for it.. So, each second more data could be transferred rather than how fast data is being transferred.
- File Copy or Peer to Peer Sharing software(s) would typically need more bandwidth.
- Any application that has to transfer huge volumes of data needs high bandwidth
Some applications like video streaming software would need both High bandwidth and low latency. On a client machine, more data would be transferred / second and transfer rate is fast. Else we would see patchy videos with lot of interruptions.
Database Mirroring:
Data Replication solutions like Synchronous Database Mirroring would need lower latency. Principal Database Server and Application are waiting synchronously for transaction to be sent to Secondary, Committed on Secondary and acknowledged by Secondary. In such situations, if data volume is less, then latency is more important than throughput.
Log Shipping:
Now if we consider case like Log Shipping, latency is not as important parameter as is bandwidth. Why? because nothing is waiting synchronously. Log Shipping in particular is Warm Standby Server solution so, in such data replication solution, throughput takes precedence over latency. Log Shipping is similar to an FTP or file copy solution.
Replication:
SQL Server transactional replication latency needs are in between Database Mirroring and Log Shipping. It has higher latency needs than log shipping but lower than Database Mirroring. Throughput requirements are defined by number of transactions multiplied by Size of Transactions (Volume of data to be transferred).
Log Shipping is most appropriate for situations that are constrained by network. With log shipping we could use native SQL Servers backup compression or introduce a customized compression component to compress transaction log backup before copying to secondary. This would also reduce bandwidth requirements.
Again there is one good article that details how to compute Throughput for TCP.
http://bradhedlund.com/2008/12/19/how-to-calculate-tcp-throughput-for-long-distance-links/
Thanks for reading and in next post we would complete remaining parameters..
-Guru