Cassini: A Tool for Ethernet based Storage Area Network Design
Faculty : Tzi-cker Chiueh
Member(s) : Srikant Sharma,
Shibiao Lin, &
Gang Peng, &
Ningning Zhu
Introduction
With the advent of Gigabit Ethernet, the initial concerns of bandwidth
and latency requirements of SANs no longer remain the core issue.
There are several features of Ethernet that are highly desirable
for SAN technology. The scalability of Ethernet makes it an easier
task to upgrade Ethernet SANs. The mainstream network deployment in an
enterprise is usually Ethernet based. If the SAN is also Ethernet based,
the interoperability issues of SAN and enterprise network are easily
addressed. In addition, the pervasive nature and the economies of scale
position it as a formidable challenger of Fibre Channel in the SAN arena.
The ultimate enabler of Ethernet based SAN is the iSCSI
protocol which defines semantics for block level SCSI I/O over any
IP network. iSCSI is the FCP counterpart on Ethernet networks which
maps the SCSI command set to the TCP/IP stack. The increase in the
momentum of iSCSI is evident from its availability through several major
operating system vendors and the availability of iSCSI enabled host bus
adapters from major hardware vendors. Even though it is debatable whether
iSCSI/Ethernet will completely replace Fibre Channel SAN or not, the
increasing momentum of Ethernet based SANs is undeniable.
Motivation
Although, the overall understanding of Ethernet SAN design methodology is
rapidly evolving, some of the issues are paid relatively less attention
than they fairly deserve. For instance, the issue of network capacity
planning and fabric design is a relatively ignored issue. Because of lack
of clear guidelines, most SAN designers resort to manual fabric design.
In manual design, visualizing and understanding the impact of spanning
tree based switching is hard to gage. This forces SAN designers to
over-provision network resources and aggregate multiple hosts on a fewer
number of network switches. For small SANs this may not pose any problem,
but for large deployments, over-provisioning is clearly not a desired
option. Further, aggregation cost of networks rapidly increases with
size. The impact of network switch failure in an aggregated deployment
is considerably high because of high number of affected hosts. The manual
design process requires initial deployment and subsequent improvisations
relying on the experience of the designers. SAN designers can greatly
benefit from automated tools which can precisely estimate the amount
of physical resources required and how to organize them for efficient
utilizations. Such a tool would cut down equipment cost and time to
deploy networks resulting in significant revenue savings.
Objective
We intend to explore the issues associated with the development of
tools that can devise an efficient and appropriate network topology
in an automated manner. The key to such an automation is the initial
topology configuration and the network refinement technique. We plan to
investigate how network refinement can be carried out so that a desired
low cost and efficient topology can be converged upon. We intend to
explore how different initial topologies can affect the convergence
process and how an appropriate initial topology can be devised. The
topolgy design process also has to pay close attention to the requirements
of path provisioning which stem from in-network replication.
In-network replication is higly desirable for efficient utilization of network
resources during data replication and mirroring in SANs.
Some of the challenges in using Ethernet for SAN stem from the inherent
spanning tree limitations of Ethernet and inability to provide link-layer
multicast which is often desirable to support in-network replication.
To address these problems we are looking into traffic engineering in Ethernet
SANs.
Cassini is such a tool which is aimed at automated Ethernet based
Storage Area Network architecture design.
Status
Currently Cassini is in design phase.