| Network Working Group | Y. Cheng |
| Internet-Draft | China Unicom |
| Intended status: Informational | C. Zhou |
| Expires: March 21, 2015 | Huawei Technologies |
| G. Karagiannis | |
| University of Twente | |
| JF. Tremblay | |
| Viagenie | |
| September 17, 2014 |
Use Cases for Distributed Data Center Applicatinos in SUPA
draft-cheng-supa-ddc-use-cases-00
This document provides several distributed data center (ddc) use cases and explains how an operator could use SUPA (Shared Unified Policy Automation) to provide these applications.
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 21, 2015.
Copyright (c) 2014 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.
The SUPA (Shared Unified Policy Automation) work aims at providing the network management application-based policy protocol(s), mechanisms and models required by network management applications to easily, accurately, and efficiently select and use the available communication network capabilities through the use of network management policies. A Network Management Application is used by an a communications service provider and/or operator to monitor, control, analyze and manage a communication network. An example of a Network Management Application is a set of actions used by an Operational Support System (OSS) entity to perform network configuration. Several SUPA use cases have been introduced in the problem statement document. This document reviews various use cases for Distributed Data Center (DDC) applications.
Take a large-scale Internet Data Center (IDC) operator as an example, it provides server hosting, bandwidth, value-added services to enterprises and ISPs, and has more than 10 data centers using over one Tbps of bandwidth in a capital city. In this IDC network, traffic at each site is routed via configuring policy routes and adjusting routes prioritization to choose an outgoing link. This type of static provisioning comes with high costs and poor operability. Furthermore, the link bandwidth resources in the data centers are not efficiently utilized.
Services usually do not have consistent bandwidth requirements at all times of the day, e.g. a video service provider usually requires less bandwidth during business hours and more during evenings. Some applications have relative high QoS requirements that may change over time., For example provisioning bandwidth and QoS for all clients of an Instant Messaging (IM) app is not reasonable and not a cost-effective solution. The operator would like to be able to optimize traffic routes dynamically so as to have the ability to load balance between data centers and links, and direct customer traffic via policies (e.g., models, software programs routines) based on customer grade and QoS requirements.It will also be useful to monitor the real-time traffic flow and have a visualized report.
Traffic engineering applications can provide dynamic traffic adjustment demands to the network based on link statuss reported by the network.
SUPA will define network management application-based policy protocol(s), mechanisms and models required to map application's demands to network management policies en procedures (e.g. traffic redirection based on customer's grade and link status), which can be directly enforced by a network management system on network devices, to meet the operator's demands.
This document illustrates several distributed datacenter (DDC) applications and explains how an operator could use SUPA to provide these applications.
The terminology used in the SUPA problem statement draft [ID.karagiannis-aponf-problem-statement-00] applies also to this draft.
There are many challenges in traditional data centers: 1) Framework and bandwidth is mainly leased, depending on manual planning and design, which leads to low resource usage and high cost; 2)Service expansion is limited in single physical DC. Each DC resource is isolated, so service and resource can only be deployed in one single DC; 3)VAS (Value Added Service) is provisioned via static configuration, which brings complex draining, long service TTM time and bad flexibility. This could not meet the requirements of complex use cases, e.g., too many VAS devices, big difference of service requirements.
To solve the above challenges for data center ISPs, SUPA could be applied in the following ways:
The following sections will illustrate four typical cases in distributed data center which could benefit from SUPA architecture. The policies transmitted from Network Management Applications to the DCs will be represented by a network configuration model described in [ID.adel-supa-configuration-model].
A large-scale data center may have more than one hundred links. The network between data centers is often leased and the applied bandwidth is very expensive. if the traditional shortest path algorithm is used to calculate a path based on static cost, then the path calculation cannot be dynamically adjusted based on real-time bandwidth usage. This will result in bandwidth waste.
Figure 1 shows how to improve the bandwidth usage efficiency beween data centers. There are two paths from DC A to DC B, for example, A-->B (path 1) and A-->C-->B (path 2). When the bandwidth between A and B is not sufficient, A will automatically transmit the traffic via C. The network management applications will configure a threshold T (e.g., 80%) for the path bandwidth usage ratio and send it to A. When an application request is received, A will detect the bandwidth usage of both paths. When the bandwidth usage ratio of path 1 (T1) has exceeded value T (e.g., 90%), while the bandwidth usage ratio of path 2 (T2) is much less than T (e.g., 10%), it will transmit the traffic to B via C, even though P1 is the shortest path between A and B.
In this case, the available bandwidth between A and B will be used efficiently, and risks of congestion between the datacenters will be avoided.
+-------------------+
|Network Management |
| |
|Applications |
+--------+----------+
| +----------+
Policy | | |
(Threshold,T) | -> B |
| / | |
| T1 / +----^-----+
| / |
+---v-----+ / |
| |/ |
| A + | T2
| |\ |
+---------+ \ |
\ |
T2 \ +----+-----+
\ | |
-> C |
| |
+----------+
Figure 1: Bandwidth usage optimization for DC Interconnection
A Data center involves many systems and the server synchronization is specifically important for DCs. Once there is error in server synchronization, the system will not run regularly, which brings mistakes and failures. However, the server synchronization is not easy to be realized during the daytime when the Data Center servers are fully loaded services. Instead, many operators choose to make the synchronization in the evening at some regular intervals.
Figure 2 shows how server synchronization between datacenters can be realized. Two servers separately in DC A and DC B are required to synchronize daily. The Network Management Applications, as defined in the SUPA architecture, are configured with several Policies, e.g., syn time (2am to 3am everyday for instance) on when to synchronize the servers, BW (a required bandwidth to be maintained between the period), and the path information (which path between the two DCs costs lower). The Network Management Applications will send these policy information to both DC A and DC B. In this case, the two servers synchronize automatically everyday from 2am to 3am, which will guarantee the normal operation of the servers.
+--------------------+
|Network Management |
| |
|Applications |
| |
+---------+----------+
/ \
/ \
/ \
/ Policy \
/ (Syn Time,BW, \
/ Path) \
| |
+-----v----+ +---v------+
| | | |
| A +--------------+ B |
| | | |
+----------+ +----------+
Figure 2: Server Synchronization between DCs
Traditional routing algorithms do not consider real-time link conditions, some requirements of specific applications cannot be met timely, e.g., delay is a key requirement for the audio services (Skype for example). How to select a better link based on the delay of each link becomes important for the application.
Figure 3 shows an example of link selection between datacenters according to the delay of each link. A value "d" is configured in the Network Management Applications for the specific applications, e.g., less than 100 ms. The value "d" will be sent to the ingress data center A for the A to detect the delays in both links between A and B. A will transmit the traffic via the link 1 to B if d1 is less than d and d2 is larger than d. In this case, the service quality and QoE user experience will be enhanced.
+--------------------+
|Network Management |
| |
|Applications |
| |
+--------+-----------+
|
|
| Policy (delay value "d")
|
|
| d2
+----v----+ ------------ +----------+
| |/ \ | |
| A |-----------------> B |
| | d1 | |
+---------+ +----------+
Figure 3: Low Delay Link Selection between Datacenters
Figure 4 illustrates a problem related to bandwidth fragmentation. From DC A to DC B, two paths (A-->B, A-->C-->B) can be reached. From A to B, only 2Gbps bandwidth is left and 8Gbps is used, and from A to B via C, the link capacity is 2Gbps. So there is no bandwidth to transmit the traffic when there is a 4Gbps requirement from A to B, which causes that the bandwidth is not effectively used.
There is no
bandwidth to
+----------+ transmit 4G +----------+
| | traffic | |
| A +------------------+ B |
| |10G link (8G used,| |
+----------+ 2G left) +----------+
\ /
\ /
\2G 2G/
\ /
\ /
+----------+
| |
| C |
| |
+----------+
Figure 4: Bandwidth Fragmentation Problem
Figure 5 provides a method to create on-demand path and bundle the path capabilities between datacenters. The bandwidth bundle capability is configured and sent to the DC A by the network management applications. When the bandwidth is not sufficient to meet the requirements for a specific application, A could bundle the bandwidth in the two links. The network capability, e.g., bandwidth bundle capability, is firstly negotiated between network management applications and the network element via other methods, which are out of the scope of this document.
+--------------------+
|Network Management |
| |
|Applications |
| |
+---------+----------+
|
|Policy (Bundle two paths)
|
|
+-----v----+ +----------+
| | 2G | |
| A +--------------> B |
| | | |
+----------+ +----^-----+
\ /
\ /
\ 2G 2G /
\ /
\ /
+----------+
| |
| C |
| |
+----------+
Figure 5: On-demand Path Creation between DCs
Security is a key aspect of any protocol that allows state installation and extracting of detailed configuration states. More investigation remains to fully define the security requirements, such as authorization and authentication levels.
Not applicable.
N/A.