Press Inquiries Contact:
Lisa Frieden (for LightCounting)
(510) 525-7945
press@lightcounting.com

Analyst Relations Contact:
Vladimir G. Kozlov
LightCounting
1 (408) 962-4851
vladimir.kozlov@lightcounting.com

LightCounting News: June 2010

HP's New Server and Virtual Connect Announcements Energize Transceiver Market.

By Kimball Brown, Senior Analyst, LightCounting, LLC

On June 22, 2010, HP announced ten new ProLiant G7 servers (three rack servers and seven blade servers), and an Ethernet/Fibre Channel module that plugs into its BladeSystem that together enable the first mass volume converged fabric in a blade server. On June 29, HP announced a top-of-rack switch that enables the rack servers to be configured in the converged fabric. By converged fabric, we mean IP (i.e. LAN traffic) and storage (either Fibre Channel (FC) or iSCSI) traffic carried on the same network. The major point here is that HP is upgrading the standard pre-installed interface on its blade servers from 1GigE to 10GigE opening up a major opportunity for increased sales of SFP+ optical transceiver modules beginning in the fourth quarter of 2010 with growth throughout 2011 and beyond. HP did not design in any support for 10GBase-T or QSFP in its servers or switches.

The Servers and Switches
Five of the seven blade servers announced now have a 10GigE LAN on Motherboard (LOM) design that enables IP, FC, and iSCSI traffic all running on the two embedded 10GigE ports on each blade motherboard. The LOM is sourced from Emulex through its announced acquisition of ServerEngines. The three rack servers can all be outfitted with the same technology with an HP add-in adapter built by Emulex that has an SFP+ interface that the user can connect either an SFP+ optical module or an SFP+ Twinax copper cable. With HPs 56% share of blades and 38% share of x86 servers, the number of 10GigE ports shipping in servers later this year and next will grow exponentially. HPs converged fabric is enabled by the HP Virtual Connect FlexFabric 10-Gb/24-port module, a blade module that installs in HPs BladeSystem and interfaces with the blade backplane via 10GBase-KR and in turn to the outside world via eight SFP+ uplink modules. The blade servers do not use SFP+ modules as the interface to the blade system backplane is via 10GBase-KR. Rather the opportunity for SFP+ modules in blades is in the eight uplinks in the blade module. The blade module was co-developed by HP with QLogic.

Further, on June 29, HP announced the A5820, a top-of-rack (ToR) switch. Both the module and the switch take traffic from up to 16 blades (in a blade module) or rack servers (in a ToR switch), discern the type of packet (i.e., IP, Fibre Channel, or iSCSI), and send the appropriate packets to eight SFP+ uplinksfour 10GigE, two FC, and two BladeSystem interconnectsfor a true converged fabric coming from the servers. In turn, the FC packets are forwarded on to an existing SAN for access to storage, and the IP packets are forwarded to the LAN. Each of the connections from the rack servers to the switch and the switch uplinks use SFP+ transceivers or Twinax cables.

Accelerated Volume Heralds Growth in Transceiver Market
The HP announcements represent the companys first phase in FCoE implementation. This first phase addresses FCoE within the rack using either blade or rack servers and switches. As traffic exits the blade or ToR switches, IP and FC traffic are split and sent to existing SANs and LANs. The second phase is the implementation of multi-hop FCoE networks where the FCoE packets are sent through a converged FCoE network. That phase will take a lot more validation and FCoE switch design. The final stage will be when FCoE targets become available and are fully validated providing fully converged FCoE networks. By implementing only the first phase, HP has simplified cabling and management within the rack, but existing FC SANs and 10 GbE LANs are unaffected. Phases two and three may take several years before users are comfortable moving to them.

This first phase has some important implications for the optical transceiver market. On the one hand, the 16 blade servers use no transceivers as they connect to the BladeSystem backplane via 10GBase-KR. The uplinks, however, should see the attachment of as many as eight, SFP+ SR, LR, or LRM modules. That makes 16 servers sharing only 8 SFP+ transceivers or SFP+ Twinax copper cables. On the other hand, the BladeSystem implementation makes the converged fabric so simple to implement that LightCounting expects a large uptake in 10GigE FCoE implementation over the next 1218 months driving strong adoption of SFP+ modules as users upgrade 1GigE connections to 10GigE connections. Further, the rack server and switch announcements are also significant as each of the connections from the rack servers to the switch and the switch uplinks use SFP+ optical transceivers or Twinax cables.

In the end, LightCounting sees the integration/validation work HP announced in June as a major driver to the acceptance of 10GigE and FC/iSCSI converged fabrics going forward. HP is the big gorilla in the high-volume server space, and the low cost of 10GigE LOMs (10s of dollars) compared to add-in adapters (100s of dollars) combined with the relative simplicity and high performance of the new switches should appeal to their customers. The big unknown is whether end users will adopt the new SFP+ Twinax cables in lieu of SFP+ optical modules and cables for short distances (i.e. 1-5 meters). End users know and trust optical networks, but the decidedly lower cost of short connections that the Twinax cables provide may become too tempting to pass up. Later this year, LightCounting will be updating our forecast of the transceiver market to reflect these trends.