White Paper

8GFC Deployment
© 2008 Silverton Consulting, Inc. 3
All Rights Reserved
P.O. Box 492|Broomfield, CO 80038-0492
USA|[phone]
concurrent streams may need as high as 600 MB/s of I/O bandwidth to
sustain adequate streaming performance.
In the above examples, only the video streaming application is a good candidate for
8GFC host connectivity. The other applications have low bandwidth requirements and
are unlikely to require 8GFC.
These published application requirements are further substantiated at least as to decision
support by Transaction Processing Performance Council (TPC) benchmark results
analysis. This analysis indicates that TPC-H (TPC’s decision support benchmark) can
attain storage bandwidth requirements approaching 350 MB/s under maximum server
transaction processing. Under more normal operation conditions much lower bandwidth
requirements would be expected.
On the other hand, application workloads aren’t the only thing driving data center
bandwidth requirements. Storage backups are often some of the worst I/O bandwidth
users today. Presently with snapshot technology, which allows for near-instantaneous
replication of data on the storage subsystem, the backup window is no longer such a
critical data center metric but storage capacity and usage continues to increase.
Consequently, more and more storage needs to be backed up for each cycle, consuming
significant bandwidth to complete in a timely fashion. For instance, a recently published
Microsoft Exchange Solution Review Program (ESRP) result for the EMC Symmetrix
DMX-4 4500, a high-end storage subsystem supporting 60,000 mailboxes, reported an
aggregate database backup throughput of ~3.9 GB/s using 14 Exchange servers each of
which averaged approximately 280 MB/s of storage throughput. As such, this high-end
storage subsystem could easily use five or more 8GFC links while each Exchange server
could still use one 4GFC for performance or two 4GFC links for availability.
8GFC and storage subsystems
Demand for increased performance of the entire IT department’s infrastructure has been
felt even at the storage subsystem level. With the move to 8GFC by server and storage
area networking links, storage connection link migration to 8GFC is expected. However,
the move is still justifiable only by a need for additional bandwidth above 4GFC.
A relatively new benchmark from the Storage Performance Consortium (SPC) reveals
that approximately one-third of storage subsystems evaluated can exceed storage
throughput of 800 MB/s. For example, one benchmark result for the IBM DS8300
indicates about a 3.2 GB/s storage throughput for the subsystem. This bandwidth usage
could justify the deployment of four or more 8GFC links.
Even though a perusal of benchmark results would suggest that few subsystems could
effectively use 8GFC links, an individual data center may benefit from additional
bandwidth provided by 8GFC. For example, many switch vendors suggest a storage link
fan-in of 5 or more to 1, i.e., 5 or more host fibre channel links per 1 storage subsystem
link. However, fan-in ratios can be much higher than 5:1 depending on the applications