Replication between 2 x DL4300 appliances over iSCSI

Hi Richard:
If I understand correctly, you would like to share the nics used for iSCSI with the replication traffic.

It is possible albeit it is not a recommended setup.

There are a few things to take into consideration.

First, is the topology of the connection between the 2 cores. More specifically, is it a local connection or does it connect geographically distinct sites?

In both cases you need to check what is the end to end sustainable bandwidth. For instance you won't take advantage of your nics capabilities if the two sites are connected by a 50Mb/s link.

Second, the most important performance factor is given by the IOPS the storage subsystem is able to yield. For instance, as an extreme example, 1Gb/s connection won't help if the storage is located on an USB 2.0 drive.

Third, it is important to asses the load on your cores. If most of the active time is performed doing internal IOPS intensive operations, for instance rollups, deferred deletes and mountability checks, the overall performance will suffer as well.

A real life example is the customer who had a very large environment with great specs, and used a 64GB deduplication cache. Since the dedupe cache was flushed on disk every hour and it was located on the C:\ volume (RAID 1), the repository was idle about 30 minutes per hour (and all core activities, including transfers, were paused) due to flushing the cache. Moving the cache on a small SSD RAID array (which was already available in his environment!) improved significantly the overall performance.

Hope that this helps.

Hi Richard:
If I understand correctly, you would like to share the nics used for iSCSI with the replication traffic.

It is possible albeit it is not a recommended setup.

There are a few things to take into consideration.

First, is the topology of the connection between the 2 cores. More specifically, is it a local connection or does it connect geographically distinct sites?

In both cases you need to check what is the end to end sustainable bandwidth. For instance you won't take advantage of your nics capabilities if the two sites are connected by a 50Mb/s link.

Second, the most important performance factor is given by the IOPS the storage subsystem is able to yield. For instance, as an extreme example, 1Gb/s connection won't help if the storage is located on an USB 2.0 drive.

Third, it is important to asses the load on your cores. If most of the active time is performed doing internal IOPS intensive operations, for instance rollups, deferred deletes and mountability checks, the overall performance will suffer as well.

A real life example is the customer who had a very large environment with great specs, and used a 64GB deduplication cache. Since the dedupe cache was flushed on disk every hour and it was located on the C:\ volume (RAID 1), the repository was idle about 30 minutes per hour (and all core activities, including transfers, were paused) due to flushing the cache. Moving the cache on a small SSD RAID array (which was already available in his environment!) improved significantly the overall performance.

Hope that this helps.