Server virtualization: The case for iSCSI

joey

Server virtualization: The case for iSCSI By Dave Simpson
In the context of server virtualization and storage, end users and vendors
agree: Separate the storage from the server. Maximizing the benefits of server
virtualization (such as resource consolidation) requires shared storage, which
means SANs.
According to International Data Corp., about 80% of virtual servers are
connected to SANs. And, today, virtually all of them are Fibre Channel SANs.
However, iSCSI-based IP SANs may have inherent advantages in the context of
server virtualization environments.

According to Matt Baker, product manager, storage marketing, at Dell, the
benefits of iSCSI in a virtual server environment fall into three categories:

• Reducing the complexity and costs associated with shared storage;
  
• Facilitating virtual machine (VM) mobility, which is one of the key value
  propositions of server virtualization; and
  
• Improving data protection, such as backups and disaster recovery.

“In many cases, iSCSI provides a superior fabric for server virtualization
compared to Fibre Channel, not just a cheaper one,” says Baker.
Praveen Asthana, Dell’s marketing director, takes it one step further:
“Server virtualization is a ‘killer app’ for iSCSI.”
Reduced complexity
iSCSI’s ability to reduce the complexity and costs associated with SANs is
not an advantage that’s specific to virtual server environments. But reduced
complexity is particularly important in virtual server environments because it
feeds into the value proposition of consolidation and simplified management. In
addition, many small and medium-sized companies embarking on server
virtualization do not have Fibre Channel expertise, nor do they have installed
Fibre Channel SANs.
In addition to reduced complexity, iSCSI lowers the entry costs for shared
storage in virtualized environments because it’s based on Ethernet, and
companies can leverage less-expensive (compared to Fibre Channel) equipment and
existing skills.
VM
mobility
With a shared-storage SAN on the back-end of a virtual server environment, if
one server goes down, the guest operating system (OS) and applications will
transfer to another physical server automatically, usually without any
disruption noticeable to users. This mobility of virtual machines and their
applications is a key benefit of server virtualization, and SANs are required
for mobility.
VM mobility also provides the ability to move workloads around to dynamically
level out (load-balance) resources, providing applications with more horsepower
on-demand.
With direct-attached storage (DAS), in contrast, if a VM fails or becomes
overloaded, administrators have to manually migrate virtual machines and
applications. SANs facilitate mobility, and iSCSI may provide some mobility
advantages compared to Fibre Channel that are, again, related to complexity.
“Fibre Channel is a very physically oriented protocol,” explains Dell’s
Baker. “WWNs are like MAC addresses: They’re burned into the hardware [e.g.,
host bus adapters]. There is no logical equivalent of WWNs that you can give to
a virtual machine, which means you have to create relationships upon
relationships between storage and virtual platforms, and then you have to again
allocate storage from the hypervisor up to the VMs.”
As such, using Fibre Channel in a virtual server environment increases the
number of “touch points” (manual configuration steps) required to manage your
storage, and transferring from a virtual environment to a physical machine over
Fibre Channel can require extensive migration planning and reconfiguration,
according to Baker.
Also, to facilitate VM mobility, fabric zoning and masking must be opened up,
so that each virtual server has access to storage. To the guest operating
systems, the provisioned storage looks as if it is directly connected, but the
guest OS does not have a direct relationship to the storage.
Initiatives such as N_Port ID Virtualization (NPIV), which allows multiple
Fibre Channel initiators to share a single physical port with multiple WWNs, may
help simplify the configuration and management of Fibre Channel SANs in
virtualized environments. However, NPIV can add fabric complexity and cost.

In contrast, Baker argues, iSCSI is very logically (as opposed to physically)
oriented (see figure, p. 19). It runs on top of Ethernet, IP, and TCP, which
gives users the ability to abstract away from the hardware and deal with the
storage configuration in a more logical way. For example, users can create a
logical one-to-one relationship among VMs, applications, and storage; this is in
contrast to the multiple “touch points” that you have to deal with in the case
of Fibre Channel. And iSCSI initiators are “agnostic” to lower-level (physical)
layers, allowing a direct relationship between a guest operation system’s
software initiator and the storage resources. As such, provisioning storage
through the VM hypervisor layer (e.g., via ESX) is no longer necessary.

“The IQN [iSCSI Qualified Name, or identifier] is tied directly to the VM,
which simplifies things by reducing the complexity of the relationship among a
VM, its applications, and storage,” says Baker. “iSCSI makes it easier to
configure the virtual environment.”
Dell refers to the ability to access storage directly from a VM without
interference from the underlying VM as “Storage Direct” (see figure, above).
Baker rounds out the case for iSCSI with comments on potential benefits of
data protection:

• With iSCSI, you can perform direct backups to tape or disk from a guest OS
  (virtual machine). With Fibre Channel, in contrast, backups have to be managed
  and arbitrated in the VM hypervisor (e.g., via ESX in the case of VMware).
  
• With iSCSI, backups (and other storage management applications) are directly
  managed in the guest OS, with direct access to storage and full application
  functionality. With Fibre Channel, backups are managed by the guest OS and
  hypervisor, which arbitrate access to storage and control the relationship with
  the external disk array. This approach can also limit the functionality of the
  applications. In an iSCSI implementation, integration with VSS/VDS can be ported
  over directly from existing backup methodologies. This is not possible with the
  indirect relationship required with Fibre Channel, according to Baker.
  “Utilization of existing backup scripts and methodologies is money in the bank
  for IT administrators,” he says. “Moreover, you can achieve much finer grained
  backup capabilities.”
  
• With iSCSI, images and applications developed on guest operating systems can
  be migrated to a non-virtualized (physical) server seamlessly. Migrating from
  virtual to physical machines via Fibre Channel can require significant
  reconfiguration by administrators and comes with the risk that administrator
  errors will cause application-level problems. The same holds true with
  physical-to-virtual and virtual-to-virtual migrations.

Nevertheless, Fibre Channel still has two advantages: It’s more mature and,
in almost all cases (or at least until 10Gbps iSCSI takes off), Fibre Channel
provides better performance. However, in the majority of virtual server
applications, iSCSI SANs may provide sufficient performance.
Chris Poelker, vice president of enterprise solutions at FalconStor Software,
cites many of the same benefits of iSCSI in virtual server environments as does
Dell’s Baker, most notably in the areas of lower cost, simplicity, disaster
recovery, and direct storage connections to virtual machines. But Poelker adds
that “performance is actually another place where iSCSI can shine,” citing not
only the advent of 10Gbps Ethernet, but also InfiniBand.
“In larger organizations we see a migration toward leveraging iSCSI as a
protocol over InfiniBand, which runs at 20Gbps, to provide RDMA access to disk,”
says Poelker.
“So in a virtualized, large-scale grid environment using a single InfiniBand
connection, you can run Fibre Channel, Ethernet, and iSCSI RDMA, which allows
you to transfer to disk at 20Gbps. Although iSCSI was originally pushed to the
back burner for performance reasons, it’s now being used for higher-performance
applications,” adds Poelker.
InfoStor July, 2007
Author(s)
:   Dave Simpson

本文来自ChinaUnix博客，如果查看原文请点：http://blog.chinaunix.net/u/2671/showart_359650.html