Network Working Group B. Constantine Internet-Draft JDSU Intended status: Informational G. Forget Expires: March 24, 2011 Bell Canada (Ext. Consultant) Reinhard Schrage Schrage Consulting September 24, 2010 Framework for TCP Throughput Testing draft-ietf-ippm-tcp-throughput-tm-07.txt Abstract This document describes a framework for measuring sustained TCP throughput performance in an end-to-end managed network environment. This document is intended to provide a practical methodology to help users validate the TCP layer performance of a managed network, which should provide a better indication of end-user experience. In the framework, various TCP and network parameters are identified that should be tested as part of the network verification at the TCP layer. Requirements Language The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119]. Status of this Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on March 24, 2011. Constantine, et al. Expires March 24, 2011 [Page 1] Internet-Draft Framework for TCP Throughput Testing August 2010 Copyright Notice Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1 Test Set-up and Terminology . . . . . . . . . . . . . . . 4 2. Scope and Goals of this methodology. . . . . . . . . . . . . . 5 2.1 TCP Equilibrium State Throughput . . . . . . . . . . . . . 6 2.2 Metrics for TCP Throughput Tests . . . . . . . . . . . . . 7 3. TCP Throughput Testing Methodology . . . . . . . . . . . . . . 9 3.1 Determine Network Path MTU . . . . . . . . . . . . . . . . 11 3.2. Baseline Round Trip Time and Bandwidth . . . . . . . . . . 13 3.2.1 Techniques to Measure Round Trip Time . . . . . . . . 13 3.2.2 Techniques to Measure End-end Bandwidth . . . . . . . 14 3.3. TCP Throughput Tests . . . . . . . . . . . . . . . . . . . 14 3.3.1 Calculate Optimum TCP Window Size. . . . . . . . . . . 15 3.3.2 Conducting the TCP Throughput Tests. . . . . . . . . . 17 3.3.3 Single vs. Multiple TCP Connection Testing . . . . . . 18 3.3.4 Interpretation of the TCP Throughput Results . . . . . 19 3.4. Traffic Management Tests . . . . . . . . . . . . . . . . . 19 3.4.1 Traffic Shaping Tests. . . . . . . . . . . . . . . . . 20 3.4.1.1 Interpretation of Traffic Shaping Test Results. . . 20 3.4.2 RED Tests. . . . . . . . . . . . . . . . . . . . . . . 21 3.4.2.1 Interpretation of RED Results . . . . . . . . . . . 21 4. Security Considerations . . . . . . . . . . . . . . . . . . . 22 5. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22 5.1. Registry Specification . . . . . . . . . . . . . . . . . . 22 5.2. Registry Contents . . . . . . . . . . . . . . . . . . . . 22 6. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 22 7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 22 7.1 Normative References . . . . . . . . . . . . . . . . . . . 22 7.2 Informative References . . . . . . . . . . . . . . . . . . 23 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 23 Constantine, et al. Expires March 24, 2011 [Page 2] Internet-Draft Framework for TCP Throughput Testing August 2010 1. Introduction Network providers are coming to the realization that Layer 2/3 testing and TCP layer testing are required to more adequately ensure end-user satisfaction. Testing an operational network prior to customer activation is referred to as "turn-up" testing and the SLA (Service Level Agreement) is generally based upon Layer 2/3 information rate, packet delay, loss and delay variation. Therefore, the network provider community desires to measure network throughput performance at the TCP layer. Measuring TCP throughput provides a meaningful measure with respect to the end user experience (and ultimately reach some level of TCP testing interoperability which does not exist today). Additionally, end-users (business enterprises) seek to conduct repeatable TCP throughput tests between enterprise locations. Since these enterprises rely on the networks of the providers, a common test methodology (and metrics) would be equally beneficial to both parties. So the intent behind this TCP throughput methodology is to define a methodology for testing sustained TCP layer performance. In this document, sustained TCP throughput is that amount of data per unit time that TCP transports during equilibrium (steady state), i.e. after the initial slow start phase. We refer to this state as TCP Equilibrium, and that the equilibrium throughput is the maximum achievable for the TCP connection(s). There are many variables to consider when conducting a TCP throughput test and this methodology focuses on some of the most common parameters that MUST be considered such as: - Path MTU and Maximum Segment Size (MSS) - RTT and Bottleneck BW - Ideal TCP Window (Bandwidth Delay Product) - Single Connection and Multiple Connection testing This methodology proposes a test which SHOULD be performed in addition to traditional Layer 2/3 type tests, which are conducted to verify the integrity of the network before conducting TCP tests. Examples include iperf (UDP mode) or manual packet layer test techniques where packet throughput, loss, and delay measurements are conducted. When available, standardized testing similar to RFC 2544 [RFC2544] but adapted for use on operational networks may be used (because RFC 2544 methods are not intended for use outside the lab environment). Constantine, et al. Expires March 24, 2011 [Page 3] Internet-Draft Framework for TCP Throughput Testing August 2010 1.1 Test Set-up and Terminology This section provides a general overview of the test configuration for this methodology. The test is intended to be conducted on an end-end operational network, so there are multitudes of network architectures and topologies that can be tested. This test set-up diagram is very general and the main intent is to illustrate the segmentation of the end user and network provider domains. Common terminologies used in the test methodology are: - Customer Provided Equipment (CPE), refers to customer owned - Customer Edge (CE), refers to provider owned demarcation device - Provider Edge (PE), refers to provider located distribution equipment - P (Provider), refers to provider core network equipment - Bottleneck Bandwidth*, lowest bandwidth along the complete network path - Round-Trip Time (RTT), refers to Layer 4 back and forth delay - Round-Trip Delay (RTD), refers to Layer 1 back and forth delay - Network Under Test (NUT), refers to the tested IP network path - TCP Throughput Test Device (TCP TTD), refers to compliant TCP host that generates traffic and measures metrics as defined in this methodology +----+ +----+ +----+ +----+ +---+ +---+ +----+ +----+ +----+ +----+ | | | | | | | | | | | | | | | | | | | | | TCP|-| CPE|-| CE |--| PE |-| P |--| P |-| PE |--| CE |-| CPE|-| TCP| | TD | | | | |BB| | | | | | | |BB| | | | | TD | +----+ +----+ +----+**+----+ +---+ +---+ +----+**+----+ +----+ +----+ <------------------------ NUT ------------------------> <-------------------------RTT ------------------------> * Bottleneck Bandwidth and Bandwidth are used synonomously in this document. ** Most of the time the Bottleneck Bandwidth is in the access portion of the wide area network (CE - PE) Note that the NUT may consist of a variety of devices including (and NOT limited to): load balancers, proxy servers, WAN acceleration devices. The detailed topology of the NUT MUST be considered when conducting the TCP throughput tests, but this methodology makes no attempt to characterize TCP performance related to specific network architectures. Constantine et al. Expires March 24, 2011 [Page 4] Internet-Draft Framework for TCP Throughput Testing August 2010 2. Scope and Goals of this Methodology Before defining the goals of this methodology, it is important to clearly define the areas that are out-of-scope for this methodology. - The methodology is not intended to predict TCP throughput behavior during the transient stages of a TCP connection, such as initial slow start. - The methodology is not intended to definitively benchmark TCP implementations of one OS to another, although some users MAY find some value in conducting qualitative experiments. - The methodology is not intended to provide detailed diagnosis of problems within end-points or the network itself as related to non-optimal TCP performance, although a results interpretation section for each test step MAY provide insight into potential issues within the network. - The methodology does not propose a method to operate permanently with high measurement loads. TCP performance and optimization data of operational networks MAY be captured and evaluated by using data of the "TCP Extended Statistics MIB" [RFC4898]. - The methodology is not intended to measure TCP throughput as part of an SLA, or to compare the TCP performance between service providers or to compare between implementations of this methodology (test equipment). In contrast to the above exclusions, the goals of this methodology are to define a method to conduct a structured, end-to-end assessment of sustained TCP performance within a managed business class IP network. A key goal is to establish a set of "best practices" that an engineer SHOULD apply when validating the ability of a managed network to carry end-user TCP applications. The specific goals are to: - Provide a practical test approach that specifies well understood, end-user configurable TCP parameters such as TCP Window size, MSS (Maximum Segment Size), number of connections, and how these affect the outcome of TCP performance over a network. - Provide specific test conditions (link speed, RTT, TCP Window size, etc.) and maximum achievable TCP throughput under TCP Equilibrium conditions. For guideline purposes, provide examples of these test conditions and the maximum achievable TCP throughput during the equilibrium state. Section 2.1 provides specific details concerning the definition of TCP Equilibrium within the context of this methodology. Constantine, et al. Expires March 24, 2011 [Page 5] Internet-Draft Framework for TCP Throughput Testing August 2010 - Define three (3) basic metrics that can be used to compare the performance of TCP connections under various network conditions. - In test situations where the RECOMMENDED procedure does not yield the maximum achievable TCP throughput result, this methodology provides some possible areas within the end host or network that SHOULD be considered for investigation (although again, this methodology is not intended to provide a detailed diagnosis of these issues). 2.1 TCP Equilibrium State Throughput TCP connections have three (3) fundamental congestion window phases as documented in [RFC5681]. These phases are: - Slow Start, which occurs during the beginning of a TCP transmission or after a retransmission time out event. - Congestion avoidance, which is the phase during which TCP ramps up to establish the maximum attainable throughput on an end-end network path. Retransmissions are a natural by-product of the TCP congestion avoidance algorithm as it seeks to achieve maximum throughput on the network path. - Retransmission phase, which include Fast Retransmit (Tahoe) and Fast Recovery (Reno and New Reno). When a packet is lost, the Congestion avoidance phase transitions to a Fast Retransmission or Recovery Phase dependent upon the TCP implementation. The following diagram depicts these phases. | ssthresh TCP | | Through- | | Equilibrium put | |\ /\/\/\/\/\ Retransmit /\/\ ... | | \ / | Time-out / | | \ / | _______ _/ | Slow _/ |/ | / | Slow _/ | Start _/ Congestion |/ |Start_/ Congestion | _/ Avoidance Loss | _/ Avoidance | _/ Event | _/ | _/ |/ |/__________________________________________________________ Time Constantine, et al. Expires March 24, 2011 [Page 6] Internet-Draft Framework for TCP Throughput Testing August 2010 This TCP methodology provides guidelines to measure the equilibrium throughput which refers to the maximum sustained rate obtained by congestion avoidance before packet loss conditions occur (which MAY cause the state change from congestion avoidance to a retransmission phase). All maximum achievable throughputs specified in Section 3 are with respect to this equilibrium state. 2.2 Metrics for TCP Throughput Tests This framework focuses on a TCP throughput methodology and also provides several basic metrics to compare results of various throughput tests. It is recognized that the complexity and unpredictability of TCP makes it impossible to develop a complete set of metrics that account for the myriad of variables (i.e. RTT variation, loss conditions, TCP implementation, etc.). However, these basic metrics will facilitate TCP throughput comparisons under varying network conditions and between network traffic management techniques. The first metric is the TCP Transfer Time, which is simply the measured time it takes to transfer a block of data across simultaneous TCP connections. The concept is useful when benchmarking traffic management techniques, where multiple connections MAY be REQUIRED. The TCP Transfer time MAY also be used to provide a normalized ratio of the actual TCP Transfer Time versus Ideal Transfer Time. This ratio is called the TCP Transfer Index and is defined as: Actual TCP Transfer Time ------------------------- Ideal TCP Transfer Time The Ideal TCP Transfer time is derived from the network path bottleneck bandwidth and the various Layer 1/2/3 overheads associated with the network path. Additionally, the TCP Window size must be tuned to equal the bandwidth delay product (BDP) as described in Section 3.3.1. The following table illustrates a single connection TCP Transfer and the Ideal TCP Transfer time for a 100 MB file with the ideal TCP window size based on the BDP. Constantine, et al. Expires March 24, 2011 [Page 7] Internet-Draft Framework for TCP Throughput Testing August 2010 Table 2.2: Link Speed, RTT, TCP Throughput, Ideal TCP Transfer time Link Maximum Achievable Ideal TCP Transfer time Speed RTT (ms) TCP Throughput(Mbps) Time in seconds -------------------------------------------------------------------- T1 20 1.17 684.93 T1 50 1.40 570.61 T1 100 1.40 570.61 T3 10 42.05 19.03 T3 15 42.05 19.03 T3 25 41.52 18.82 T3(ATM) 10 36.50 21.92 T3(ATM) 15 36.23 22.14 T3(ATM) 25 36.27 22.05 100M 1 91.98 8.70 100M 2 93.44 8.56 100M 5 93.44 8.56 1Gig 0.1 919.82 0.87 1Gig 0.5 934.47 0.86 1Gig 1 934.47 0.86 10Gig 0.05 9,344.67 0.09 10Gig 0.3 9,344.67 0.09 * Calculation is based on File Size in Bytes X 8 / TCP Throughput. ** TCP Throughput is derived from Table 3.3. To illustrate the TCP Transfer Time Index, an example would be the bulk transfer of 100 MB over 5 simultaneous TCP connections (each connection uploading 100 MB). In this example, the Ethernet service provides a Committed Access Rate (CAR) of 500 Mbit/s. Each connection MAY achieve different throughputs during a test and the overall throughput rate is not always easy to determine (especially as the number of connections increases). The ideal TCP Transfer Time would be ~8 seconds, but in this example, the actual TCP Transfer Time was 12 seconds. The TCP Transfer Index would be 12/8 = 1.5, which indicates that the transfer across all connections took 1.5 times longer than the ideal. The second metric is the TCP Efficiency metric which is the percentage of bytes that were not retransmitted and is defined as: Transmitted Bytes - Retransmitted Bytes --------------------------------------- x 100 Transmitted Bytes Transmitted bytes are the total number of TCP payload bytes to be transmitted which includes the original and retransmitted bytes. This metric provides a comparative measure between various QoS mechanisms such as traffic management, congestion avoidance, and also various TCP implementations (i.e. Reno, Vegas, etc.). Constantine, et al. Expires March 24, 2011 [Page 8] Internet-Draft Framework for TCP Throughput Testing August 2010 As an example, if 100,000 bytes were sent and 2,000 had to be retransmitted, the TCP Efficiency SHOULD be calculated as: 102,000 - 2,000 ---------------- x 100 = 98.03% 102,000 Note that the retransmitted bytes MAY have occurred more than once, and these multiple retransmissions are added to the Retransmitted Bytes count (and the Transmitted Bytes count). And the third metric is the Buffer Delay Percentage, which represents the increase in RTT during a TCP throughput test from the inherent network RTT (baseline RTT). The baseline RTT is the round-trip time inherent to the network path under non-congested conditions. The Buffer Delay Percentage is defined as: Average RTT during Transfer - Baseline RTT ------------------------------------------ x 100 Baseline RTT As an example, the baseline RTT for the network path is 25 msec. During the course of a TCP transfer, the average RTT across the entire transfer increased to 32 msec. In this example, the Buffer Delay Percentage WOULD be calculated as: 32 - 25 ------- x 100 = 28% 25 Note that the TCP Transfer Time, TCP Efficiency, and Buffer Delay metrics MUST be measured during each throughput test. Poor TCP Transfer Time Indexes (TCP Transfer Time greater than Ideal TCP Transfer Times) MAY be diagnosed by correlating with sub-optimal TCP Efficiency and/or Buffer Delay Percentage metrics. 3. TCP Throughput Testing Methodology As stated in Section 1, it is considered best practice to verify the integrity of the network by conducting Layer2/3 stress tests such as [RFC2544] or other methods of network stress tests. If the network is not performing properly in terms of packet loss, jitter, etc. then the TCP layer testing will not be meaningful since the equilibrium throughput MAY be very difficult to achieve (in a "dysfunctional" network). Constantine, et al. Expires March 24, 2011 [Page 9] Internet-Draft Framework for TCP Throughput Testing August 2010 TCP Throughput testing MAY require cooperation between the end user customer and the network provider. In a Layer 2/3 VPN architecture, the testing SHOULD be conducted on the Customer Edge (CE) router and not the Provider Edge (PE) router. The following represents the sequential order of steps to conduct the TCP throughput testing methodology: 1. Identify the Path MTU. Packetization Layer Path MTU Discovery or PLPMTUD, [RFC4821], MUST be conducted to verify the maximum network path MTU. Conducting PLPMTUD establishes the upper limit for the MSS to be used in subsequent steps. 2. Baseline Round Trip Time and Bandwidth. This step establishes the inherent, non-congested Round Trip Time (RTT) and the bottleneck bandwidth of the end-end network path. These measurements are used to provide estimates of the ideal TCP window size, which SHOULD be used in subsequent test steps. These measurements reference [RFC2681] and [RFC4898] to measure RTD (and the associated RTT). Also, [RFC5136] is referenced to measure network capacity. 3. TCP Connection Throughput Tests. With baseline measurements of Round Trip Time and bottleneck bandwidth, a series of single and multiple TCP connection throughput tests SHOULD be conducted to baseline the network performance expectations. 4. Traffic Management Tests. Various traffic management and queuing techniques SHOULD be tested in this step, using multiple TCP connections. Multiple connection testing SHOULD verify that the network is configured properly for traffic shaping versus policing, various queuing implementations, and RED. Important to note are some of the key characteristics and considerations for the TCP test instrument. The test host MAY be a standard computer or dedicated communications test instrument and these TCP test hosts be capable of emulating both a client and a server. Whether the TCP test host is a standard computer or a compliant TCP TTD, the following areas SHOULD be considered when selecting a test host: - TCP implementation used by the test host OS version, i.e. Linux OS kernel using TCP Reno, TCP options supported, etc. This will obviously be more important when using custom test equipment where the TCP implementation MAY be customized or tuned to run in higher performance hardware. When a compliant TCP TTD is used, the TCP implementation SHOULD be identified in the test results. The compliant TCP TTD SHOULD be usable for complete end-to-end testing through network security elements and SHOULD also be usable for testing network sections. Constantine, et al. Expires March 24, 2011 [Page 10] Internet-Draft Framework for TCP Throughput Testing August 2010 - Most importantly, the TCP test host must be capable of generating and receiving stateful TCP test traffic at the full link speed of the network under test. As a general rule of thumb, testing TCP throughput at rates greater than 100 Mbit/sec MAY require high performance server hardware or dedicated hardware based test tools. Thus, other devices cannot realize higher TCP throughput, and user expectations SHOULD be set accordingly with user manual or notes on the results report. - Measuring RTT and TCP Efficiency per connection will generally require dedicated hardware based test tools. In the absence of dedicated hardware based test tools, these measurements MAY need to be conducted with packet capture tools (conduct TCP throughput tests and analyze RTT and retransmission results with packet captures). Another option MAY be to use "TCP Extended Statistics MIB" per [RFC4898]. - The compliant TCP TTD and its access to the network under test MUST NOT introduce a performance bottleneck of any kind. 3.1. Determine Network Path MTU TCP implementations SHOULD use Path MTU Discovery techniques (PMTUD). PMTUD relies on ICMP 'need to frag' messages to learn the path MTU. When a device has a packet to send which has the Don't Fragment (DF) bit in the IP header set and the packet is larger than the Maximum Transmission Unit (MTU) of the next hop link, the packet is dropped and the device sends an ICMP 'need to frag' message back to the host that originated the packet. The ICMP 'need to frag' message includes the next hop MTU which PMTUD uses to tune the TCP Maximum Segment Size (MSS). Unfortunately, because many network managers completely disable ICMP, this technique does not always prove reliable in real world situations. Packetization Layer Path MTU Discovery or PLPMTUD [RFC4821] MUST be conducted to verify the minimum network path MTU. PLPMTUD can be used with or without ICMP. The following sections provide a summary of the PLPMTUD approach and an example using the TCP protocol. [RFC4821] specifies a search_high and a search_low parameter for the MTU. As specified in [RFC4821], a value of 1024 is a generally safe value to choose for search_low in modern networks. It is important to determine the overhead of the links in the path, and then to select a TCP MSS size corresponding to the Layer 3 MTU. For example, if the MTU is 1024 bytes and the TCP/IP headers are 40 bytes, then the MSS would be set to 984 bytes. Constantine, et al. Expires March 24, 2011 [Page 11] Internet-Draft Framework for TCP Throughput Testing August 2010 An example scenario is a network where the actual path MTU is 1240 bytes. The TCP client probe MUST be capable of setting the MSS for the probe packets and could start at MSS = 984 (which corresponds to an MTU size of 1024 bytes). The TCP client probe would open a TCP connection and advertise the MSS as 984. Note that the client probe MUST generate these packets with the DF bit set. The TCP client probe then sends test traffic per a nominal window size (8KB, etc.). The window size SHOULD be kept small to minimize the possibility of congesting the network, which MAY induce congestive loss. The duration of the test should also be short (10-30 seconds), again to minimize congestive effects during the test. In the example of a 1240 byte path MTU, probing with an MSS equal to 984 would yield a successful probe and the test client packets would be successfully transferred to the test server. Also note that the test client MUST verify that the MSS advertised is indeed negotiated. Network devices with built-in Layer 4 capabilities can intercede during the connection establishment process and reduce the advertised MSS to avoid fragmentation. This is certainly a desirable feature from a network perspective, but can yield erroneous test results if the client test probe does not confirm the negotiated MSS. The next test probe would use the search_high value and this would be set to MSS = 1460 to correspond to a 1500 byte MTU. In this example, the test client MUST retransmit based upon time-outs (since no ACKs will be received from the test server). This test probe is marked as a conclusive failure if none of the test packets are ACK'ed. If any of the test packets are ACK'ed, congestive network MAY be the cause and the test probe is not conclusive. Re-testing at other times of the day is RECOMMENDED to further isolate. The test is repeated until the desired granularity of the MTU is discovered. The method can yield precise results at the expense of probing time. One approach MAY be to reduce the probe size to half between the unsuccessful search_high and successful search_low value, and increase by increments of 1/2 when seeking the upper limit. Constantine, et al. Expires March 24, 2011 [Page 12] Internet-Draft Framework for TCP Throughput Testing August 2010 3.2. Baseline Round Trip Time and Bandwidth Before stateful TCP testing can begin, it is important to determine the baseline Round Trip Time (non-congested inherent delay) and bottleneck bandwidth of the end-end network to be tested. These measurements are used to provide estimates of the ideal TCP window size, which SHOULD be used in subsequent test steps. These latency and bandwidth tests SHOULD be run during the time of day for which the TCP throughput tests will occur. 3.2.1 Techniques to Measure Round Trip Time Following the definitions used in the section 1.1; Round Trip Time (RTT) is the time elapsed between the clocking in of the first bit of a payload packet to the receipt of the last bit of the corresponding Acknowledgment. Round Trip Delay (RTD) is used synonymously to twice the Link Latency. RTT measurements SHOULD use techniques defined in [RFC2681] or statistics available from MIBs defined in [RFC4898]. The RTT SHOULD be baselined during "off-peak" hours to obtain a reliable figure for inherent network latency versus additional delay caused by network buffering delays. During the actual sustained TCP throughput tests, RTT MUST be measured along with TCP throughput. Buffer delay effects can be isolated if RTT is concurrently measured. This is not meant to provide an exhaustive list, but summarizes some of the more common ways to determine round trip time (RTT) through the network. The desired resolution of the measurement (i.e. msec versus usec) may dictate whether the RTT measurement can be achieved with standard tools such as ICMP ping techniques or whether specialized test equipment would be required with high precision timers. The objective in this section is to list several techniques in order of decreasing accuracy. - Use test equipment on each end of the network, "looping" the far-end tester so that a packet stream can be measured end-end. This test equipment RTT measurement MAY be compatible with delay measurement protocols specified in [RFC5357]. - Conduct packet captures of TCP test applications using for example "iperf" or FTP, etc. By running multiple experiments, the packet captures can be studied to estimate RTT based upon the SYN -> SYN-ACK handshakes within the TCP connection set-up. Constantine, et al. Expires March 24, 2011 [Page 13] Internet-Draft Framework for TCP Throughput Testing August 2010 - ICMP Pings MAY also be adequate to provide round trip time estimations. Some limitations of ICMP Ping MAY include msec resolution and whether the network elements respond to pings (or block them). 3.2.2 Techniques to Measure End-end Bandwidth There are many well established techniques available to provide estimated measures of bandwidth over a network. This measurement SHOULD be conducted in both directions of the network, especially for access networks which MAY be asymmetrical. Measurements SHOULD use network capacity techniques defined in [RFC5136]. The bandwidth measurement test MUST be run with stateless IP streams (not stateful TCP) in order to determine the available bandwidth in each direction. And this test SHOULD obviously be performed at various intervals throughout a business day (or even across a week). Ideally, the bandwidth test SHOULD produce a log output of the bandwidth achieved across the test interval. 3.3. TCP Throughput Tests This methodology specifically defines TCP throughput techniques to verify sustained TCP performance in a managed business network. Defined in section 2.1, the equilibrium throughput reflects the maximum rate achieved by a TCP connection within the congestion avoidance phase on an end-end network path. This section and others will define the method to conduct these sustained throughput tests and guidelines of the predicted results. With baseline measurements of round trip time and bandwidth from section 3.2, a series of single and multiple TCP connection throughput tests can be conducted to baseline network performance against expectations. It is RECOMMENDED to run the tests in each direction independently first, then run both directions simultaneously. In each case, the TCP Transfer Time, TCP Efficiency, and Buffer Delay metrics MUST be measured in each direction. Constantine, et al. Expires March 24, 2011 [Page 14] Internet-Draft Framework for TCP Throughput Testing August 2010 3.3.1 Calculate Ideal TCP Window Size The ideal TCP window size can be calculated from the bandwidth delay product (BDP), which is: BDP (bits) = RTT (sec) x Bandwidth (bps) By dividing the BDP by 8, the "ideal" TCP window size is calculated. An example would be a T3 link with 25 msec RTT. The BDP would equal ~1,105,000 bits and the ideal TCP window would equal ~138,000 bytes. The following table provides some representative network link speeds, latency, BDP, and associated ideal TCP window size. Sustained TCP transfers SHOULD reach nearly 100% throughput, minus the overhead of Layers 1-3 and the divisor of the MSS into the TCP Window. For this single connection baseline test, the MSS size will effect the achieved throughput (especially for smaller TCP Window sizes). Table 3.2 provides the achievable, equilibrium TCP throughput (at Layer 4) using 1460 byte MSS. Also in this table, the 58 byte L1-L4 overhead including the Ethernet CRC32 is used for simplicity. Table 3.3: Link Speed, RTT and calculated BDP, TCP Throughput Link Ideal TCP Maximum Achievable Speed* RTT (ms) BDP (bits) Window (kBytes) TCP Throughput(Mbps) --------------------------------------------------------------------- T1 20 30,720 3.84 1.17 T1 50 76,800 9.60 1.40 T1 100 153,600 19.20 1.40 T3 10 442,100 55.26 42.05 T3 15 663,150 82.89 42.05 T3 25 1,105,250 138.16 41.52 T3(ATM) 10 407,040 50.88 36.50 T3(ATM) 15 610,560 76.32 36.23 T3(ATM) 25 1,017,600 127.20 36.27 100M 1 100,000 12.50 91.98 100M 2 200,000 25.00 93.44 100M 5 500,000 62.50 93.44 1Gig 0.1 100,000 12.50 919.82 1Gig 0.5 500,000 62.50 934.47 1Gig 1 1,000,000 125.00 934.47 10Gig 0.05 500,000 62.50 9,344.67 10Gig 0.3 3,000,000 375.00 9,344.67 * Note that link speed is the bottleneck bandwidth for the NUT Constantine, et al. Expires March 24, 2011 [Page 15] Internet-Draft Framework for TCP Throughput Testing August 2010 Also, the following link speeds (available payload bandwidth) were used for the WAN entries: - T1 = 1.536 Mbits/sec (B8ZS line encoding facility) - T3 = 44.21 Mbits/sec (C-Bit Framing) - T3(ATM) = 36.86 Mbits/sec (C-Bit Framing & PLCP, 96000 Cells per second) The calculation method used in this document is a 3 step process : 1 - Determine what SHOULD be the optimal TCP Window size value based on the optimal quantity of "in-flight" octets discovered by the BDP calculation. We take into consideration that the TCP Window size has to be an exact multiple value of the MSS. 2 - Calculate the achievable layer 2 throughput by multiplying the value determined in step 1 with the MSS & (MSS + L2 + L3 + L4 Overheads) divided by the RTT. 3 - Finally, multiply the calculated value of step 2 by the MSS versus (MSS + L2 + L3 + L4 Overheads) ratio. This provides the achievable TCP Throughput value. Sometimes, the maximum achievable throughput is limited by the maximum achievable quantity of Ethernet Frames per second on the physical media. Then this value is used in step 2 instead of the calculated one. The following diagram compares achievable TCP throughputs on a T3 link with Windows 2000/XP TCP window sizes of 16KB versus 64KB. 45| | _____42.1M 40| |64K| TCP | | | Throughput 35| | | _____34.3M in Mbps | | | |64K| 30| | | | | | | | | | 25| | | | | | | | | | 20| | | | | _____20.5M | | | | | |64K| 15| 14.5M____| | | | | | | |16K| | | | | | 10| | | | 9.6M+---+ | | | | | | | |16K| | 5.8M____+ | 5| | | | | | | |16K| | |______+___+___+_______+___+___+_______+__ +___+_______ 10 15 25 RTT in milliseconds Constantine, et al. Expires March 24, 2011 [Page 16] Internet-Draft Framework for TCP Throughput Testing August 2010 The following diagram shows the achievable TCP throughput on a 25ms T3 when the TCP Window size is increased and with the [RFC1323] TCP Window scaling option. 45| | +-----+42.47M 40| | | TCP | | | Throughput 35| | | in Mbps | | | 30| | | | | | 25| | | | ______ 21.23M | | 20| | | | | | | | | | 15| | | | | | | | | | 10| +----+10.62M | | | | | _______5.31M | | | | | | 5| | | | | | | | | |__+_____+______+____+__________+____+________+_____+___ 16 32 64 128 TCP Window size in KBytes 3.3.2 Conducting the TCP Throughput Tests There are several TCP tools that are commonly used in the network world and one of the most common is the "iperf" tool. With this tool, hosts are installed at each end of the network segment; one as client and the other as server. The TCP Window size of both the client and the server can be manually set and the achieved throughput is measured, either uni-directionally or bi-directionally. For higher BDP situations in lossy networks (long fat networks or satellite links, etc.), TCP options such as Selective Acknowledgment SHOULD be considered and also become part of the window size / throughput characterization. Host hardware performance MUST be well understood before conducting the TCP throughput tests and other tests in the following sections. Dedicated test equipment will generally be REQUIRED, especially for line rates of GigE and 10 GigE. A compliant TCP TTD SHOULD provide a warning message when the expected test throughput will exceed 10% of the network bandwidth capacity. If the throughput test is expected to exceed 10% of the provider bandwidth, then the test SHOULD be coordinated with the network provider. This does not include the customer premise bandwidth, the 10% refers directly to the provider's bandwidth (Provider Edge to Provider router). Constantine, et al. Expires March 24, 2011 [Page 17] Internet-Draft Framework for TCP Throughput Testing August 2010 The TCP throughput test SHOULD be run over a long enough duration to properly exercise network buffers and also characterize performance during different time periods of the day. Note that both the TCP Transfer Time, TCP Efficiency, and Buffer Delay metrics MUST be measured during each throughput test. Poor TCP Transfer Time Indexes (TCP Transfer Time greater than Ideal TCP Transfer Times) MAY be diagnosed by correlating with sub-optimal TCP Efficiency and/or Buffer Delay Percentage metrics. 3.3.3 Single vs. Multiple TCP Connection Testing The decision whether to conduct single or multiple TCP connection tests depends upon the size of the BDP in relation to the window sizes configured in the end-user environment. For example, if the BDP for a long-fat pipe turns out to be 2MB, then it is probably more realistic to test this pipe with multiple connections. Assuming typical host computer window settings of 64 KB, using 32 connections would realistically test this pipe. The following table is provided to illustrate the relationship of the BDP, window size, and the number of connections required to utilize the available capacity. For this example, the network bandwidth is 500 Mbps, RTT is equal to 5 ms, and the BDP equates to 312 KBytes. #Connections Window to Fill Link ------------------------ 16KB 20 32KB 10 64KB 5 128KB 3 The TCP Transfer Time metric is useful for conducting multiple connection tests. Each connection SHOULD be configured to transfer payloads of the same size (i.e. 100 MB), and the TCP Transfer time SHOULD provide a simple metric to verify the actual versus expected results. Note that the TCP transfer time is the time for all connections to complete the transfer of the configured payload size. From the example table listed above, the 64KB window is considered. Each of the 5 connections would be configured to transfer 100MB, and each TCP should obtain a maximum of 100 Mb/sec per connection. So for this example, the 100MB payload should be transferred across the connections in approximately 8 seconds (which would be the ideal TCP transfer time for these conditions). Additionally, the TCP Efficiency metric SHOULD be computed for each connection tested (defined in section 2.2). Constantine, et al. Expires March 24, 2011 [Page 18] Internet-Draft Framework for TCP Throughput Testing August 2010 3.3.4 Interpretation of the TCP Throughput Results At the end of this step, the user will document the theoretical BDP and a set of Window size experiments with measured TCP throughput for each TCP window size setting. For cases where the sustained TCP throughput does not equal the ideal value, some possible causes are listed: - Network congestion causing packet loss which MAY be inferred from a poor TCP Efficiency metric (100% = no loss) - Network congestion causing an increase in RTT which MAY be inferred from the Buffer Delay metric (0% = no increase in RTT over baseline) - Intermediate network devices which actively regenerate the TCP connection and can alter window size, MSS, etc. - Rate limiting (policing). More discussion of traffic management tests follows in section 3.4 3.4. Traffic Management Tests In most cases, the network connection between two geographic locations (branch offices, etc.) is lower than the network connection of the host computers. An example would be LAN connectivity of GigE and WAN connectivity of 100 Mbps. The WAN connectivity may be physically 100 Mbps or logically 100 Mbps (over a GigE WAN connection). In the later case, rate limiting is used to provide the WAN bandwidth per the SLA. Traffic management techniques are employed to provide various forms of QoS, the more common include: - Traffic Shaping - Priority queuing - Random Early Discard (RED, etc.) Configuring the end-end network with these various traffic management mechanisms is a complex under-taking. For traffic shaping and RED techniques, the end goal is to provide better performance for bursty traffic such as TCP (RED is specifically intended for TCP). This section of the methodology provides guidelines to test traffic shaping and RED implementations. As in section 3.3, host hardware performance MUST be well understood before conducting the traffic shaping and RED tests. Dedicated test equipment will generally be REQUIRED for line rates of GigE and 10 GigE. If the throughput test is expected to exceed 10% of the provider bandwidth, then the test SHOULD be coordinated with the network provider. This does not include the customer premise bandwidth, the 10% refers directly to the provider's bandwidth (Provider Edge to Provider router). Constantine, et al. Expires March 24, 2011 [Page 19] Internet-Draft Framework for TCP Throughput Testing August 2010 3.4.1 Traffic Shaping Tests For services where the available bandwidth is rate limited, there are two (2) techniques used to implement rate limiting: traffic policing and traffic shaping. Simply stated, traffic policing marks and/or drops packets which exceed the SLA bandwidth (in most cases, excess traffic is dropped). Traffic shaping employs the use of queues to smooth the bursty traffic and then send out within the SLA bandwidth limit (without dropping packets unless the traffic shaping queue is exceeded). Traffic shaping is generally configured for TCP data services and can provide improved TCP performance since the retransmissions are reduced, which in turn optimizes TCP throughput for the given available bandwidth. Through this section, the available rate-limited bandwidth shall be referred to as the "bottleneck bandwidth". The ability to detect proper traffic shaping is more easily diagnosed when conducting a multiple TCP connection test. Proper shaping will provide a fair distribution of the available bottleneck bandwidth, while traffic policing will not. The traffic shaping tests are built upon the concepts of multiple connection testing as defined in section 3.3.3. Calculating the BDP for the bottleneck bandwidth is first REQUIRED before selecting the number of connections and TCP Window size per connection. Similar to the example in section 3.3, a typical test scenario might be: GigE LAN with a 100Mbps bottleneck bandwidth (rate limited logical interface), and 5 msec RTT. This would require five (5) TCP connections of 64 KB window size evenly fill the bottleneck bandwidth (about 100 Mbps per connection). The traffic shaping test SHOULD be run over a long enough duration to properly exercise network buffers (greater than 30 seconds) and also characterize performance during different time periods of the day. The throughput of each connection MUST be logged during the entire test, along with the TCP Transfer Time, TCP Efficiency, and Buffer Delay metrics. 3.4.1.1 Interpretation of Traffic Shaping Test Results By plotting the throughput achieved by each TCP connection, the fair sharing of the bandwidth is generally very obvious when traffic shaping is properly configured for the bottleneck interface. For the previous example of 5 connections sharing 500 Mbps, each connection would consume ~100 Mbps with a smooth variation. If traffic policing was present on the bottleneck interface, the bandwidth sharing MAY not be fair and the resulting throughput plot MAY reveal "spikey" throughput consumption of the competing TCP connections (due to the retransmissions). Constantine, et al. Expires March 24, 2011 [Page 20] Internet-Draft Framework for TCP Throughput Testing August 2010 3.4.2 RED Tests Random Early Discard techniques are specifically targeted to provide congestion avoidance for TCP traffic. Before the network element queue "fills" and enters the tail drop state, RED drops packets at configurable queue depth thresholds. This action causes TCP connections to back-off which helps to prevent tail drop, which in turn helps to prevent global TCP synchronization. Again, rate limited interfaces can benefit greatly from RED based techniques. Without RED, TCP is generally not able to achieve the full bandwidth of the bottleneck interface. With RED enabled, TCP congestion avoidance throttles the connections on the higher speed interface (i.e. LAN) and can reach equilibrium with the bottleneck bandwidth (achieving closer to full throughput). The ability to detect proper RED configuration is more easily diagnosed when conducting a multiple TCP connection test. Multiple TCP connections provide the multiple bursty sources that emulate the real-world conditions for which RED was intended. The RED tests also build upon the concepts of multiple connection testing as defined in section 3.3.3. Calculating the BDP for the bottleneck bandwidth is first REQUIRED before selecting the number of connections and TCP Window size per connection. For RED testing, the desired effect is to cause the TCP connections to burst beyond the bottleneck bandwidth so that queue drops will occur. Using the same example from section 3.4.1 (traffic shaping), the 500 Mbps bottleneck bandwidth requires 5 TCP connections (with window size of 64Kb) to fill the capacity. Some experimentation is REQUIRED, but it is RECOMMENDED to start with double the number of connections to stress the network element buffers / queues. In this example, 10 connections SHOULD produce TCP bursts of 64KB for each connection. If the timing of the TCP tester permits, these TCP bursts SHOULD stress queue sizes in the 512KB range. Again experimentation will be REQUIRED and the proper number of TCP connections and TCP window size will be dictated by the size the network element queue. 3.4.2.1 Interpretation of RED Results The default queuing technique for most network devices is FIFO based. Without RED, the FIFO based queue will cause excessive loss to all of the TCP connections and in the worst case global TCP synchronization. Constantine, et al. Expires March 24, 2011 [Page 21] Internet-Draft Framework for TCP Throughput Testing August 2010 By plotting the aggregate throughput achieved on the bottleneck interface, proper RED operation MAY be determined if the bottleneck bandwidth is fully utilized. For the previous example of 10 connections (window = 64 KB) sharing 500 Mbps, each connection SHOULD consume ~50 Mbps. If RED was not properly enabled on the interface, then the TCP connections will retransmit at a higher rate and the net effect is that the bottleneck bandwidth is not fully utilized. Another means to study non-RED versus RED implementation is to use the TCP Transfer Time metric for all of the connections. In this example, a 100 MB payload transfer SHOULD take ideally 16 seconds across all 10 connections (with RED enabled). With RED not enabled, the throughput across the bottleneck bandwidth MAY be greatly reduced (generally 20-40%) and the actual TCP Transfer time MAY be proportionally longer then the Ideal TCP Transfer time. Additionally, the TCP Transfer Efficiency metric is useful, since non-RED implementations MAY exhibit a lower TCP Transfer Efficiency. 4. Security Considerations The security considerations that apply to any active measurement of live networks are relevant here as well. See [RFC4656] and [RFC5357]. 5. IANA Considerations This document does not REQUIRE an IANA registration for ports dedicated to the TCP testing described in this document. 6. Acknowledgments Thanks to Matt Mathis, Matt Zekauskas, Al Morton, Rudi Geib, and Yaakov Stein, Loki Jorgenson for many good comments and for pointing us to great sources of information pertaining to past works in the TCP capacity area. 7. References 7.1 Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC4656] Shalunov, S., Teitelbaum, B., Karp, A., Boote, J., and M. Zekauskas, "A One-way Active Measurement Protocol (OWAMP)", RFC 4656, September 2006. [RFC5681] Allman, M., Paxson, V., Stevens W., "TCP Congestion Control", RFC 5681, September 2009. [RFC2544] Bradner, S., McQuaid, J., "Benchmarking Methodology for Network Interconnect Devices", RFC 2544, June 1999 Constantine, et al. Expires March 24, 2011 [Page 22] Internet-Draft Framework for TCP Throughput Testing August 2010 [RFC5357] Hedayat, K., Krzanowski, R., Morton, A., Yum, K., Babiarz, J., "A Two-Way Active Measurement Protocol (TWAMP)", RFC 5357, October 2008 [RFC4821] Mathis, M., Heffner, J., "Packetization Layer Path MTU Discovery", RFC 4821, June 2007 draft-ietf-ippm-btc-cap-00.txt Allman, M., "A Bulk Transfer Capacity Methodology for Cooperating Hosts", August 2001 [RFC2681] Almes G., Kalidindi S., Zekauskas, M., "A Round-trip Delay Metric for IPPM", RFC 2681, September, 1999 [RFC4898] Mathis, M., Heffner, J., Raghunarayan, R., "TCP Extended Statistics MIB", May 2007 [RFC5136] Chimento P., Ishac, J., "Defining Network Capacity", February 2008 [RFC1323] Jacobson, V., Braden, R., Borman D., "TCP Extensions for High Performance", May 1992 7.2. Informative References Authors' Addresses Barry Constantine JDSU, Test and Measurement Division One Milesone Center Court Germantown, MD 20876-7100 USA Phone: +1 240 404 2227 barry.constantine@jdsu.com Gilles Forget Independent Consultant to Bell Canada. 308, rue de Monaco, St-Eustache Qc. CANADA, Postal Code : J7P-4T5 Phone: (514) 895-8212 gilles.forget@sympatico.ca Reinhard Schrage Schrage Consulting Phone: +49 (0) 5137 909540 reinhard@schrageconsult.com Constantine, et al. Expires March 24, 2011 [Page 23]