Filetype PDF | Posted on 08 Oct 2022 | 3 years ago
Authentication
digital resources
116x Filetype PDF File size 0.93 MB Source: www.intel.com
IT@Intel White Paper
Intel IT
IT Best Practices
Data Center Solutions
May 2012
Using Converged Network Adapters for
FCoE Network Unification
Executive Overview
Switching to FCoE deployment To unify local area network and storage area network infrastructure over a single
using dual-port Intel® Ethernet fabric, Intel IT evaluated the price and performance advantages of switching to a
X520 Server Adapters would Fibre Channel over Ethernet (FCoE) using converged network adapters (CNAs). In
yield a greater than 50-percent comparing a CNA solution to a two-card solution—a 10 Gigabit Ethernet network
reduction in total network interface card and a Fibre Channel (FC) host bus adapter (HBA)—we determined
costs per server rack. that CNAs provide a technically sound, cost-effective solution for network
unification in our virtualized environment.
Our evaluation revealed that switching • Allows application loads to drive nearly
to FCoE deployment using the dual-port 90-percent full bi-directional bandwidth
10-gigabit (Gb) Intel® Ethernet Server Adapter utilization on FCoE or TCP/IP
X520 series had the following benefits: • Provides an open solution that, taking
• Reduces total network costs per server advantage of native FCoE initiators in
rack by more than 50 percent operating systems, offers an efficient
• Delivers FC performance comparable and cost-effective alternative to more
to and, in some cases, exceeding the expensive proprietary CNAs that offload
performance of an FC HBA FC and FCoE processing onto the adapter
• Enables controllable prioritization for To realize these advantages, Intel IT plans
LAN or SAN traffic through network to use FCoE configurations and dual-port
Quality of Service mechanisms that 10-Gb Intel Ethernet X520 server adapters
regulate contention for bandwidth in new and replacement servers in our
Craig Pierce virtualized environment.
System Engineer,
Intel Architecture Systems Integration (IASI)
Sanjay Rungta
Senior Principal Engineer, Intel IT
Sreeram Sammeta
System Engineer, IASI
Terry Yoshii
Research Staff, Intel IT
IT@Intel White Paper Using Converged Network Adapters for FCoE Network Unification
Contents BACkGrOuNd 1GbE network infrastructure and found it
For years, many organizations have run inadequate to meet Intel’s rapidly growing
Executive Overview ............................. 1 business requirements and the increasing
two separate, parallel networks in their demands they place on data center resources.1
Background ............................................ 2 data center. To connect servers and clients,
How FCoE Works ............................... 2 as well as connect to the Internet, they Among the trends that supported our transition
use an Ethernet-based local area network to a 10GbE data center fabric design are:
Solution ................................................. 3 (lAN). For connecting servers to the • The escalating data handling demands
Converged Network Adapter storage area network (SAN) and the block created by increasing compute density
Test Goals and Challenges .............. 4 storage arrays used for storing data, they in our Design computing domain and by
Storage Performance use a Fibre Channel (FC)-based network. large-scale virtualization in our Office and
Comparison: Throughput
................. 4 Enterprise computing domain
Storage Performance In late 2010, Intel IT decided our existing • The need to match network performance
Comparison: Response Time ........... 5 1-gigabit Ethernet (1GbE) network demands with the increasing gains in
Storage Performance infrastructure was no longer adequate to meet file server performance due to the latest
Comparison: CPU Effectiveness ..... 5 Intel’s rapidly growing business requirements high-performance Intel® processors and
Baseline Bandwidth Utilization and the resource demands of our increasingly clustering technologies. The network, not
Quality-of-Service Testing .............. 5 virtualized environment. We needed a more the file servers, was the limiting factor in
Quality-of-Service Testing .............. 6 cost-effective solution that unifies our fabric, supporting faster throughput.
Storage I/O Latency during provides equal or better performance, and • A 40-percent annual growth in Internet
Quality-of-Service Testing .............. 7 enables traffic prioritization through quality
of service (QoS) mechanisms. Our solution: a connection requirements driving a need for
Conclusion .............................................. 7 10GbE infrastructure which we deployed initially greater bandwidth through the organization
Acronyms ................................................ 8 for connection to the local area network (LAN) When we found we could reduce our network
and to host bus adapters (HBAs) that connect total cost of ownership by as much as 18
to our FC-based storage area network (SAN). to 25 percent, we began the transition to a
The recent development of Fibre Channel 10GbE data center fabric design in 2011.
over Ethernet (FCoE), a storage protocol that In our initial implementation, we used 10GbE
enables FC communications to run directly NICs to connect servers to the LAN. To
over Ethernet, provides a way for companies connect to the FC-based SAN, we used HBAs.
to consolidate these networks into a single At the time, we viewed this as a temporary
common network infrastructure. By unifying solution until we could fully evaluate the use
LANs and SANs, eliminating redundant of converged network adapters (CNAs) to
switches, and reducing cabling and network enable a unified network fabric through FCoE.
interface card (NIC) counts, an FCoE server
adapter—specifically, the dual-port 10-Gb How FCoE Works
IT@INTEl Intel® Ethernet X520 Server Adapter—can: FCoE is essentially an extension of FC over a
• Reduce capital expenditures (CapEx) different link layer transport. FCoE maps FC
The IT@Intel program connects IT over Ethernet while remaining independent of
professionals around the world with their • Cut power and cooling costs by reducing the the Ethernet forwarding scheme. This enables
peers inside our organization – sharing number of components FC to use 10GbE networks while preserving
lessons learned, methods and strategies. • Simplify administration by reducing the the FC protocol. With the new data center
Our goal is simple: Share Intel IT best number of devices that need to be managed
practices that create business value and bridging (DCB) specification, the 10-Gb DCB
make IT a competitive advantage. Visit In addition, network unification using FCoE protocol also alleviates two previous concerns:
us today at www.intel.com/IT or contact reduces operating expenditures. For a large IT packet loss and latency.
your local Intel representative if you’d organization such as Intel IT, this is a significant Servers connect to FCoE with CNAs, which
like to learn more. advantage. Our 87 data centers support a combine both FC HBA and Ethernet NIC
massive worldwide computing environment
1
that houses approximately 90,000 servers. We address in detail our work evaluating the
advantages to Intel of upgrading to 10GbE
The timing of FCoE’s release is advantageous network infrastructure in the IT@Intel white paper
2 www.intel.com/IT as well. We recently evaluated Intel’s existing “Upgrading Data Center Network Architecture to
10 Gigabit Ethernet.” Intel Corporation, January 2011.
Using Converged Network Adapters for FCoE Network Unification IT@Intel White Paper
functionality on a single adapter card. Use of FCoE
and CNAs consolidates network TCP/IP and SAN data Fibre Channel over Ethernet (FCoE)
traffic on a unified network.
For Intel IT, a unified fabric using CNAs represents an One disadvantage of Fibre Channel (FC) as a network technology for storage
important next step in the ongoing effort to make the area networks (SANs) is that it’s incompatible with Ethernet, the dominant
conversion of our data center architecture to 10GbE server networking technology. Although SANs and Ethernet networks perform
connections as cost-efficient as possible. substantively the same function, they use different technologies and thus are
entirely separate physical networks, with separate switches, cabling, networking
hardware (such as network interface cards and host bus adapters), and connections
SOluTION into each server. The expertise required to support each network is also different.
We examined two approaches to providing Current industry standards combine server and storage networks over a single
a converged adapter. One approach uses an unified fabric. The most prominent of these is FCoE. This specification preserves
open solution that takes advantage of native the storage-specific benefits of FC and allows it to be transported over Ethernet
FCoE initiators in operating systems, enabling in the enterprise. Under FCoE, both the Ethernet and the FC protocols are
the adapter to work in a complementary way merged, enabling organizations to deploy a single, converged infrastructure
with the platform hardware and the operating carrying both SAN and server TCP/IP network traffic. FCoE preserves all FC
system. This approach provides a cost-effective constructs, providing reliable delivery, while preserving and interoperating
way to handle both FCoE and TCP/IP traffic with an organization’s investment in FC SANs, equipment, tools, and training.
(see Figure 1).
Host Program Interface Local Area Network Configuration
FC Host Bus Adapter API, DCBx, FC File System (LAN) Data center bridging exchange (DCBx)
link protocol and fibre channel (FC)
multipath I/O iSCSI
Native Operating System or
Device Drivers Intel® FCoE Initiator and Management
virtual Storport*, Open FC, TCP/IP FCoE, encapsulation and decapsulation,
FCoE Protocol FCoE initialization protocol
Network Driver Interface
Specification, Net Device
Base Drivers Intel® Ethernet Base Drivers with
with Buffers and Interrupts Data Center Bridging
Priorities assigned to traffic classes (802.1p)
Adapter Traffic Classes and Queues
Storage Area Network LAN Traffic class mapped to queues (802.1Qaz)
Classification and Quality of Service Engine
Prioritization Engine Insert tag .1Q , schedule priority groups,
allocate bandwidth, flow control
FCoE Offloads FCoE Offloads: DDP, CRC Rx and Tx, LSO, RSS
Data path acceleration
Converged Port
FCoE Converged LAN Priority tagged packets, congestion messages
MAC Address Port MAC Address
vLAN 1 vLAN 2
API – application programming interface: CRC – cyclic redundancy check; DDP – direct data placement; FCoE – Fibre Channel over Ethernet; iSCSI – internet Small Computer System Interface;
LSO – large segment offload; MAC – media access control; RSS – really simple syndication; Rx – receive or incoming (data); Tx – transmit or outgoing (data); vLAN – virtual LAN
Figure 1. This diagram shows a unified networking Fibre Channel over Ethernet (FCoE) configuration. The converged network adapter uses native
FCoE initiators in the operating system to work with the platform hardware and operating system, handling FCoE traffic at a lower cost than a
hardware-based solution. www.intel.com/IT 3
IT@Intel White Paper Using Converged Network Adapters for FCoE Network Unification
The second approach uses custom • Be regulated through appropriate QoS
Quality of Service (QoS) hardware adapters with FC and FCoE (and mechanisms (see sidebar) that assign
in some cases TCP/IP) protocol processing higher or lower priority to either
In computer networking, QoS resource embedded in the hardware. With this kind of traffic—network or storage.
reservation control mechanisms prioritize approach, the CNA converts FCoE traffic to This is a critical capability because
different applications, users, or data flows FC packets in the hardware, and the CNA a converged network must be able
to ensure a certain level of performance. manufacturer provides its custom drivers, to protect storage traffic from any
For example, a particular bit rate, along with interfaces, and management software. other traffic. Today’s SAN networks
limits on delay, jitter, and packet-dropping The result is a more expensive adapter. provide this type of isolation through
probability and bit error rate, may be Intel IT decided to test the first approach— a separate infrastructure
guaranteed for an application that is delay- which is software driver-based—because • Serve as a cost-effective, technically
or loss-sensitive, such as storage or video. it would provide a robust, scalable, viable solution for network unification
For such applications, QoS guarantees high-performance server connectivity in our virtualized environment
are important when network capacity is solution without the expensive, custom
insufficient for the concurrent data flow. hardware. What’s more, we knew that Storage Performance
recent improvements in some of the Comparison: Throughput
latest hypervisors now enable them to The first test, which relies on a CNA using
work with FCoE traffic. This meant the a software driver for FCoE processing,
software driver-based approach could be had to realize equivalent or similar
used in virtualized environments. performance in storage processing to the
dual-port 8-Gb Fibre Channel HBAs we
Average I/O Operations Per Second Converged Network Adapter were replacing. To determine this, we ran
67% Read; 100% Random Test Goals and Challenges a series of tests using different I/O block
Fibre Channel Protocol (FCP) sizes—from 4 kilobytes (KB) to 128 KB—
Fibre Channel over Ethernet (FCoE) Recognizing cost as an important
100,000 factor associated with an adapter that through both types of adapters. We
could ultimately be used on hundreds deliberately avoided saturating the hosts
80.000 of servers, our goal was to see how a and network and storage arrays in these
ond CNA using a software driver for FCoE tests by specifying a cache-intensive
60.000 processing would work for Intel’s data load—5 megabytes (MB) working set size—
centers. To find out, we put dual-port and by driving moderately high I/O from
tions per Sec 16 workers to four logical unit numbers at
a 40.000 10-Gb Intel Ethernet X520 Server
Adapters, which support FCoE, in two a queue depth of 1 from a single host.
I/O Oper Intel® Xeon® processor 5600 series-based Tests showed negligible differences
20.000
servers and ran a variety of performance in the smaller block sizes. The Fibre
tests. We wanted to see if such a CNA Channel Protocol (FCP) performance of
0 could address the following challenges:
4 KB 8 KB 16 KB 32 KB 64 KB 128 KB the dual-port 8-Gb FC HBA was 13.1
FCP 84,841 77,741 65,110 42,440 26,710 13,510 • Provide storage performance to 2.5 percent higher than the FCoE
FCoE 75,045 70,051 63,491 51,724 34,414 18,272 (throughput, response time, performance of the dual-port Intel
Figure 2. Using a software driver to handle Fibre Channel CPU effectiveness, and latency) Ethernet X520 Server Adapter running
over Ethernet, the dual-port 10-Gb Intel® Ethernet X520 comparable to an FC HBA at 4-KB to 16-KB I/O block sizes (see
Server Adapter delivered 21.9 to 35.2 percent higher Figure 2). More significant was the
throughput running at 32-kilobyte (KB) to 128-KB block I/O • Drive the network load and the difference in the higher I/O block sizes.
sizes compared to a dual-port 8-Gb Fibre Channel host bus storage load to their peaks, when The FCoE performance of the dual-
adapter. each is run in isolation
4 www.intel.com/IT
no reviews yet
Please Login to review.
