Electronic Design

  
Reprints     Printer-Friendly    Email this Article    RSS        Font Size     What's This?


[Product Innovation]
Processor Manages Heavy Lifting With Dual Gigabit Ethernet Ports
A storage network processor takes on TCP/IP chores in hardware while accommodating storage protocols like iSCSI.

William Wong  |   ED Online ID #2466  |   July 8, 2002


IP-based storage networks need significant horsepower to handle large tcp/ip loads. Silverback Systems' iSNAP 2100 delivers this power while offloading the host processor to support more connections. This approach also allows the host processor to take on additional tasks, leaving TCP/IP (transmission control protocol/Internet protocol) and higher-level protocol matters, such as iSCSI, to the iSNAP 2100.

The iSNAP 2100 is designed for storage-area network (SAN) and network-attached storage (NAS) devices as well as servers, high-end workstations, and multiprotocol storage routers and switches. It will be especially effective in blade servers, where storage is designed to be distributed. In this case, the iSNAP 2100 would be incorporated on the blade server that's connected to a Gigabit Ethernet (GE) switch fabric instead of a conventional backplane.

IP-based Ethernet solutions are of great interest to enterprises due to the installed base of Ethernet currently used in LANs. Network managers are already familiar with the management of various Ethernet speeds and complex switch and router environments. Although storage-aware switches will be more efficient for network-based storage solutions, standard Ethernet switches can be employed as well.

Network-based storage offers significant advantages over local storage solutions, including scalability, distributed storage and computing servers, high reliability, redundancy, and high bandwidth. Centralized management and increasing storage also become easier with network-based storage. The primary disadvantage over local storage is latency.

Although relatively common, NAS devices don't approach the performance level of the iSNAP 2100. SANs also are becoming more common, and IP-based storage will have to compete with InfiniBand solutions. Yet in many scenarios, the two coexist because of the wide implementation of high-performance Ethernet networks.

Many chip-based storage solutions, including the iSNAP 2100, employ dual connections to the switch fabric. InfiniBand chips come in dual-port versions too. These provide re-dundant connections to the storage network fabric.

The dual-port nature of the iSNAP 2100 is useful for more than redundancy. Initially, servers will likely use one port to link to a storage-related network with SAN and NAS devices while the second joins with a LAN where workstations and other servers reside. NAS devices can be located on the LAN as well. In addition, the iSNAP 2100 may provide gateway support for existing and emerging storage technologies, like Fibre Channel and InfiniBand.

However, the iSNAP 2100 is much more than a dual-port GE adapter. For starters, it only handles both IP and TCP in hardware. In fact, the iSNAP 2100 completely offloads TCP/IP support from the host processor (Table 1). TCP/IP is a complex protocol that can sap significant amounts of a host's processing power without a device such as the iSNAP 2100. Even when running in full-duplex mode, the iSNAP 2100 does this for both GE ports.

High-Level Protocols: Reducing the load on the host processor for TCP/IP alone is a significant benefit, but the iSNAP 2100 doesn't stop there. It also is capable of handling higher-level protocols, like the emerging iSCSI and remote direct memory access (RDMA). These protocols are specific to storage applications versus other IP-based protocols such as H.323, a multimedia teleconferencing protocol.

The iSNAP 2100 has hardware de-signed to make its job easier. For example, the classification and encapsulation engines handle iSCSI packets (Fig. 1). The iSNAP 2100 also includes multiple processing nodes for dealing with large numbers of TCP/IP connections. The event queue management and scheduler coordinates the operation of these processing nodes.

Before moving data into main memory, the QMS and processing nodes carry out protocols. The nodes work with the control hub, which controls data movement from main memory to the PCI-X or SPI-3 interfaces. IP-based responses can also be sent back over the Ethernet links. The QMS is a state machine configured by the general-purpose processor.

Acting as a control-plane processor, the MIPS-based, general-purpose processor takes care of exceptions. It can handle some higher-level protocols, like Internet control message protocol (ICMP) and address resolution protocol (ARP).

This exceptional collection of processing power allows the iSNAP 2100 to take on even higher-level protocols, such as common Internet file system (CFIS) and network file system (NFS). While these services can be executed via a conventional network processing unit (NPU), the iSNAP 2100 has the edge because it's optimized for storage applications (Table 2).

The iSNAP 2100 handles connection setup and teardown chores. With a conventional NPU or GE adapter and host, tackling the connection process can be a significant overhead and generate bus traffic between the host. The iSNAP 2100 eliminates the traffic in addition to offloading the host processing.

Simultaneously, the iSNAP 2100 can accommodate many connections via the off-chip DDR-SDRAM. This design technique enables an iSNAP 2100-equipped blade server to use a smaller amount of memory because storage connections will be low compared to an NAS or SAN device that will support hundreds or thousands of server connections.


<-- prev. page     [1] 2     next page -->

Reprints   Printer-Friendly  Email this Article  RSS    Font Size   What's This?


  • 2008 BEST Electronic Design Winners
  • In EDA, A Year Of Mergers, Failed And Otherwise
  • 2008 BEST Electronic Design Winners
  • Engineers Rely On Internet For Product Info
  • Rochester Electronics Establishes New Design and Technology Group
  • November 17, 2008
  • Custom Sources Light Way To 22-nm IC Lithography
  • Software Turns Scopes Into Vector RF Signal Analyzers
    1) Switch-Mode ICs Promote Efficient Power Management, Part 1: Switch-Mode Fundamentals
    (1913 views today)
    2) Ubuntu Goes Embedded
    (453 views today)
    3) Parts Add Up To Home Theater PC
    (336 views today)
    4) Build A Smart Battery Charger Using A Single-Transistor Circuit
    (224 views today)
    5) The ABCs Of Fiber Bragg Gratings
    (197 views today)
    ALL TOP 20



    POST YOUR COMMENTS HERE
    Name:

    Email:
    Your Comments:

    Enter the text from the image below


    Please refresh the page if you have trouble reading this text.

    Search Electronic Design
         
      
     
    Web Seminar
    Sponsored By:
    Title: Read Pacing: A Performance Enhancing Feature of PCI Express Gen 2 Switch Devices
    Speakers: 
    Date: 07/01/08
    Register: 

    Electronic Design Europe Electronic Design China EEPN Power Electronics Auto Electronics Microwaves & RF
    Mobile Dev & Design Schematics Find Power Products Military Electronics EE Events Related Resources