DCLC Research Group Y. Shi INTERNET-DRAFT UCI Intended Status: Informational S. Duan Expires: January 22, 2015 CATR L. Deng China Mobile July 21, 2014 Test and Analysis on Forwarding Latency in Terms of Queuing Length draft-shi-dclc-latency-test-00 Abstract We test the influence of the queue length, packet size and so on to the forwarding latency of the typical access switch, which is meant to set off our efforts in studying the relationship between the queuing length and the transport performance of some latency- sensitive applications in a DC network. Status of this Memo This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. 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." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Copyright and License Notice Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved. Expires Jan 22, 2015 [Page 1] INTERNET DRAFT Jul 21, 2014 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 2 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 Objective of the Tests . . . . . . . . . . . . . . . . . . . . 3 4 Preparation of the Test Environment . . . . . . . . . . . . . . 3 5 Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 4 5.1 Test of Buffer Configuration in Switch . . . . . . . . . . . 4 5.2 Test of Forwarding Delay in Switch . . . . . . . . . . . . . 4 6 Result Analysis . . . . . . . . . . . . . . . . . . . . . . . . 5 7 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . 6 8 Security Considerations . . . . . . . . . . . . . . . . . . . . 6 9 IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 6 10 References . . . . . . . . . . . . . . . . . . . . . . . . . . 6 10.1 Normative References . . . . . . . . . . . . . . . . . . . 6 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 7 Expires Jan 22, 2015 [Page 2] INTERNET DRAFT Jul 21, 2014 1 Introduction To investigate the effective optimization mechanism of the congestion control in the internal network of the data center, we test the influence of the queuing length and packet size to the forwarding latency of an access switch. From the analysis, we try to find the relationship between the queuing length and the transport performance of some latency-sensitive application. 2 Terminology 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]. 3 Objective of the Tests As a typical interconnection network with super high bandwidth, simultaneous multipath and low end-to-end delay, data center network has been widely used in uploading from traditional applications and desktop virtualization to highly interconnected distributed applications. The uploading method can be made by providing physical resources to the application system, or it can be made by providing virtual sources(virtual machine, virtual network and virtual equipment, etc.) through the abstract of the physical resources which is done by virtualized infrastructure. It is suggested that the interior network of the data center still has problems with the delivery performance of the latency-sensitive application for the mismatch between end-to-end congestion control and the variation of delay in IP level. To investigate the effective optimization mechanism of the congestion control in the internal network of the data center, we test the influence of the queuing length, packet size and so on to the forwarding latency of the representative COTS switch, which is assumed to be served as a basis for finding the control mechanism for the latency-sensitive applications when they are doing end-to-end local transportation. 4 Preparation of the Test Environment We use an access switch as the test instrument. The measurement instrument is Spirent's TestCenter C1(SPT C1). To simulate the condition of the congestion, we use 3 switch ports in real test. 2 ports(port1 and port2)sent to the third one(port3) Expires Jan 22, 2015 [Page 3] INTERNET DRAFT Jul 21, 2014 simultaneously. The total two input flow is larger than the output port bandwidth. 5 Test Cases We tried two test cases to investigate: 1)buffer configuration of the switch and how the buffer space is shared; and 2)no-packet-loss forwarding time delay of the switch when enabling a tail drop feature on a queue with configured consumption ratio (i.e. the buffer(%)). 5.1 Test of Buffer Configuration in Switch The buffer configuration test aims to evaluate the size of the queue's length in the switch, and how different packets share the buffer allocated to a given port. Since there are commonly multiple queues for different QoS packets passing a single port, and we are focusing on queuing length rather than service differentiation, we ignore the class of QoS and the priority queue schemes in the switch. When setting the length of the buffer, we ask the switch to drop the packet when the buffer allocated to the queue meets a threshold in percentage (i.e. the buffer(%)). We set each of the input flow be 51% of the wire bandwidth to make sure that there will be a queue building (congestion) at the corresponding output port. Back-to-back test of the 3 ports is done by using RFC 2544 of SPT C1. The packet sizes are 512 and 1024. The test step is 0.1% of the bandwidth. The maximum number of received frame when the queue just meets the threshold is the real value of the length of the buffer. To verify the effectiveness of the value, we choose 50% queue and 100% queue to test separately to see the change of the maximum number of received frame. 5.2 Test of Forwarding Delay in Switch Forwarding delay test mainly inspects the relationship between the queuing at the switch and the delay. Therefore, we verified the maximum buffer queue length to drop the packet when the buffer(%) reached the threshold. We use RFC 2544 to go through the latency test of the three ports. Each input sets the flow percentage from 49.8% to 50.2% with an Expires Jan 22, 2015 [Page 4] INTERNET DRAFT Jul 21, 2014 increase of 0.2% each time. The tested frame sizes are 64, 128, 256, 512, 768, 1024, 1280 and 1518, as recommended in [RFC2544]. The buffer(%) is set to 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100% of the maximum queue length to analyze the delay time of different frame under different queue length. 6 Result Analysis We did a group of tests and got a bunch of data. The data were analyzed by univariate statistics first, and then by using linear regression model, we did the multivariate statistics. In this way, we found out that minimum delay time shown different relations with the variants under observation, while the maximum delay time and the average delay time share very similar features when we considered the packet size and the buffer(%) as the independent variable. 6.1 The Minimum Delay Observation 1: when buffer(%) is fixed, the delay increases linearly with the packet size. Observation 2: When packet size is fixed, the buffer(%) seems has no influence on the minimum delay time. This was also been proved when we did the multivariate analysis. The coefficient of the buffer(%) was 0 in the fitting line. 6.2 The Maximum/Average Delay Observation 1: Both the maximum and average delay have almost the same tendency and values. Observation 2: When buffer(%)is fixed, with increasing packet size, delay time will also increase. Seen from the data, the maximum and average delay time have a nonlinear relationship with the packet size. Besides, we noticed that when the packet size is too big or small, the maximum delay time and the average delay time are more influenced by the packet size compared with the one with moderate packet size. Observation 3: When packet size is fixed, maximum and average delay increase linearly with the buffer(%). Also, with bigger packet size, the delays are more influenced by the buffer(%) since the slot of the fitting line is higher. Compared with the minimum delay time, we come to the conclusion that average delay time is more influenced by maximum delay time, rather than the minimum delay time. Expires Jan 22, 2015 [Page 5] INTERNET DRAFT Jul 21, 2014 In the multivariate statistics, since the relationship between the delay time and the packet size is nonlinear, we tried a lot of nonlinear models, such as polynomial equations. The best fitting result so far is the combination of packet number(buffer(%)/packet size) and first order of buffer(%) and a constant number. Since the minimum delay time is only related to the packet size, we replaced the packet size with the minimum delay time in the maximum and average delay time's fitting equations, in this way both the maximum and average delay can be represented by the minimum delay time and buffer(%). 7 Future Works Later we will have more tests and try to have a clear understanding of the influence of the congestion and the reason for the congestion. The preliminary considerations are: 1) Test and analyze on other access switches and aggregate switches. 2) Test and analyze on a more complicated internetworking topology (e.g. sequence of or parallel union of multiple switches). 3) Measure the RTT for TCP data flow, and analyze the relationship between the forwarding delay and TCP RTT. 8 Security Considerations N/A. 9 IANA Considerations N/A. 10 References 10.1 Normative References Expires Jan 22, 2015 [Page 6] INTERNET DRAFT Jul 21, 2014 Authors' Addresses Yang Shi University of California, Irvine Email: shiy4@uci.edu Shihui Duan China Academy of Telecommunication Research of MIIT Email: duanshihui@catr.cn Lingli Deng China Mobile Email: denglingli@chinamobile.com Expires Jan 22, 2015 [Page 7]