vSphere 4.1 Storage Networking Updates

There are a lot of pieces to today’s announcement of VMware vSphere 4.1.  The biggest part of the announcement is the vStorage APIs for Array Integration (VAAI) – it is also well covered through press releases and blog posts. I’m going to capture the impact on storage networking (FC, iSCSI and FCoE).  VMware also has a reference center which includes many papers discussing the new functionality.  VMware’s full support of FC, iSCSI and FCoE is an area that differentiates it from Microsoft.  Microsoft is a strong proponent of iSCSI and does have full support of FC, but it is lacking with FCoE.

Fibre Channel: “vSphere 4.1 adds full support for 8Gb FC from the ESX server to the storage array.”  This means that companies can now take advantage of the latest generation of HBAs and storage arrays that support 8Gb FC.  While first product support from switch and adapter vendors were available in 2008, the reality is that it usually takes 12-18 months for customers to begin significant adoption of new FC speeds.  This delay is due to a mix of testing cycles, upgrade cycles and the need for the newer technology which is typically tied to newer generation servers that can support the higher bandwidth.  Server virtualization is one of the key drivers for higher bandwidth utilization, so now that VMware has support for 8Gb, customers should consider using 8Gb for any new acquisitions of HBAs, switches or storage arrays.  VMware shared some performance numbers on page 7 of this document.  Pricing between 4Gb and 8Gb is typically the same for HBAs and nominally more expensive for switches and storage arrays.

iSCSI: With vSphere 4.1, VMware adds support for “dependent” Hardware iSCSI Offload – which means that the NIC works with “VMware networking, and iSCSI configuration and management interfaces provided by VMware”.  This is especially important as we move to 10Gb Ethernet solutions which can significantly impact CPU utilization without offloads. Basically, the NIC solution provides iSCSI and other IP traffic and that has offload on the chip from a single wire.  The first solution supported by VMware is from Broadcom.  There is a lot of activity in the adapter space as vendors compete for 10Gb LOM design wins with varying functionality in offloads and protocol supports (iSCSI and FCoE).  Being supported on VMware’s HCL is a milestone for the vendors that customers should pay close attention to.

Fibre Channel over Ethernet (FCoE): VMware has been an early supported of FCoE solutions.  While not specifically part of the 4.1 announcement, VMware added support of native FCoE storage arrays since their last major release.  As stated in the introduction to the VMware SAN compatibility guide: “Native FCoE arrays connected to FCoE CNAs are also supported with ESX 4.0 and newer releases.”  Today, NetApp is the only vendor shipping native FCoE arrays.  Note that even with native FCoE, there must be an FCoE switch (from the Cisco Nexus family or Brocade 8000 family) between the server and the array.  The promise of true end-to-end Ethernet solution based on FCoE are closer to reality with the VMware support and the announced support of FCoE on Cisco’s Nexus 7000 (core Ethernet switch) which is scheduled for Q3’10.

Boot from SAN: Boot from SAN is supported with ESXi 4.1 for FC, iSCSI and FCoE solutions.  Boot support was experimental in ESXi 4.0 and earlier. As with all configurations, check the VMware HCL and release notes for specific support and configuration requirements.

My Take

VMware is doing a better job in storage networking, in particular working with its partners, than is Microsoft with Hyper-V.  As a result, the company is moving faster to become the platform of choice for mission critical applications in the datacenter.  Having said that, this is a long race and Hyper-V is not out of the picture. Specifically, VMware and its ecosystem have lots of work to do to support high value workloads such as SAP and Oracle.  There are three points I want to make in this regard:

  • An example of the ecosystem challenges is that of the new VAAI solutions, only EMC, NetApp and 3Par are ready to support in Q3.
  • Security and management continue to be weak spots in virtualization solutions.
  • Application heads are still concerned about the performance implications of migrating legacy systems.

Nonetheless, this announcement represents an important step on the journey towards virtualized data centers.

Share

, ,

  • Pingback: Tweets that mention vSphere 4.1 Storage Networking Updates « Wikibon Blog -- Topsy.com()

  • Good blog that summarizes the 4.1 Update. You are correct in that security in the virtualization solutions is a weak spot. However, EMC and Emulex have addressed this concern with the joint encryption solutions (http://bit.ly/agTC5M) that was showcased at EMC World Boston 2010.
    The Emulex OneSecure adapter is different from any other security solutions in that it allows vMotion and other virtualization duties to happen in the SAN without interference.

  • Good blog that summarizes the 4.1 Update. You are correct in that security in the virtualization solutions is a weak spot. However, EMC and Emulex have addressed this concern with the joint encryption solutions (http://bit.ly/agTC5M) that was showcased at EMC World Boston 2010.
    The Emulex OneSecure adapter is different from any other security solutions in that it allows vMotion and other virtualization duties to happen in the SAN without interference.

  • Pingback: EMC and NetApp Vie for FCoE Leadership with PR Weapons « Wikibon Blog()

  • Wikibon – I think this isn’t such a hard problem as folks make it. Networking teams need to become real NSP’s and not worry about the access l2 – and let access l2/l3s become dedicated to customers. I think that Cisco / EMC / Vmware are going to push big iron to make the l2/l3 “virtual” – but it ends up being a very very expensive wire. IMHO – FCOE is a capital play for the big guys.

    the bigger problem is that typical access layer routing is a cookie cutter approach and handing off edge routers to “customers” is scary.It would be possible for the entire edge l3 becomes an injectable entity – but the network needs to let the customer decide where to publish their l3.