Chart 14 - Concurrent Backup Of Multiple Servers Across Single 100Basetx Segment - Compaq 219700-001 - ProLiant - 1500 White Paper

Compaq Backup and Recovery for Microsoft SQL Server 6.x Page 56

Chart 14 - Concurrent Backup of Multiple Servers Across Single 100BaseTX Segment

The benefits of this concurrent strategy are obvious, as greater throughput is derived from the tape
server and more of the network bandwidth is utilized, for each additional database backed up from a
remote system. Eventually, the overall throughput reaches 23.1 GB/hr with 4 simultaneous channels of
data being transferred into the server and written out to tape, and 60% of the network bandwidth in use.
The performance increase however, is not as linear as what was observed during the concurrent
10Base-T tests. This is because all data here flows through a single point, and the CSMA/CD access
method (used by 100BaseTX) leads to frame collisions when multiple nodes are transmitting on one
segment. In the above tests, our collision rate went from 4% with 2 concurrent backups to over 21%
with 4 concurrent backups. One could isolate the data transfers onto different segments and thereby
reduce collisions by using a 100 Mbit switch in place of the repeater, by using 2 separate repeaters (one
for each NIC on the server), or by using 4 separate repeaters (along with 4 NICs on the server) as we
did for the 10 Mbit tests.
Note that Chart-14 does not include results for concurrent remote backups to 35/70-GB DLT drives.
We would expect however, that such results would be very similar to the results for the RAID-0
configured 15/30-GB DLT drives (i.e: 4 groups of a single 35/70 DLT each could be substituted in
place of the 4 groups of 2 15/30 DLT's in RAID-0), as they were in Chart-12. Remember that overall
backup throughput is limited by the rate of the "slowest link in the chain" so to speak, and with remote
backups that link is usually the network, so that a faster storage device does not always yield
performance gains.
Finally, it should be mentioned that if a single segment is used for such concurrent backup jobs, then
they should be scheduled 'after-hours' or done across a dedicated network link, so as to cause minimal
interference with other network based applications. Furthermore, the ultimate performance of these or
any other remote backup processes may vary based upon the specific physical layout of your network,