| ippm | H. Song, Ed. |
| Internet-Draft | T. Zhou |
| Intended status: Experimental | Huawei |
| Expires: December 29, 2017 | June 27, 2017 |
On Scalability of In-situ OAM
draft-song-ippm-ioam-scalability-01
This document describes several potential scalability issues when implementing in-situ OAM based on the current in-situ OAM documents and proposes the corresponding solutions and modifications to the current in-situ OAM specification. Specifically, we extend in-situ OAM to support more standard tracing data than is currently defined and add new features to avoid limitations on MTU, bandwidth, forwarding path length, and node processing capability. We provide use cases to motivate our proposal and base the changes on the current in-situ OAM header format specification.
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 December 29, 2017.
Copyright (c) 2017 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.
In-situ OAM (iOAM) records OAM information within user packets while the packets traverse a network. The data types and data formats for in-situ OAM data records have been defined in [I-D.brockners-inband-oam-data]. We identify several scalability issues for implementing the current iOAM specification and propose solutions in this draft.
Currently 11 data types and associated formats (including wide format and short format of the same data) are defined in [I-D.brockners-inband-oam-data] . The presence of data is indicated by a 16-bit bitmap in the "OAM-Trace-Type" field.
In the current specification only five bits are left to identify new data types. Moreover, some data is forced to be bundled together as a single unit to save bitmap space and pack data to the ideal size (e.g., the hop limit and the node id are bundled, and the ingress interface id and the egress interface id are bundled), regardless of the fact that an application may only ask for a part of the data. Last but not the least, each data is forced to be 4-byte aligned for easier access, resulting in waste of header space in many cases.
Since the data plane bandwidth, the data plane packet processing, and the management plane data handling are all precious yet scarce resource, the scheme should strive to be simple and precise. The application should be able to control the exact type and format of data it needs to collect and analyze. It is conceivable that more types of data may be introduced in the future. However, the current scheme cannot support it after all the bits in the bitmap are used up.
Currently, bit 7 is used to indicate the presence of variable length opaque state snapshot data. While this data field can be used to store arbitrary data, the data is difficult to be standardized and another schema is needed to decode the data, which may lead to low data plane performance.
When a flow traverses a series of middleboxes (e.g., Firewall, NAT, and load balancer), its identity (e.g., the 5-tuple) is often altered, which makes the OAM system lose track of the flow trace. In this case, we may want to copy some of the original packet header fields into the iOAM header so the original flow can be identified at any point of the network.
In wireless, mobile, and optical network environments, some physical data associated with a flow (e.g., power, temperature, signal strength, GPS location) need to be collected to monitor the service performance.
Both cases require new iOAM data types. More examples are listed in Section 3.2.
The total size of data is limited by the MTU. When the number of required data types is large and the forwarding path length is long, it is possible that there is not enough space in the iOAM header to save the data. The current proposal is to label the overflow status and stop adding new node data to the packet, leading to loss of information.
Even if the header has enough space to hold the iOAM data, the overhead may be too large and consume too much bandwidth. For example, if we assume moderate 20 bytes of data per node, a path with length of 10 will need 200 bytes to hold the data. This will inflate small 64-byte packets by more than four times. Even for the largest packet size (e.g., 1500 bytes), the overhead (>10%) is not negligible. Therefore, we need to limit the iOAM data overhead without sacrificing the data collection capability.
Here we have another interesting related issue. Packets can be dropped anywhere in a network for various reasons. If we can only collect iOAM data at the path end, we lose all data from the dropped packets and have no idea where the packets are dropped. This defies the purpose of iOAM and makes those iOAM-enabled nodes work in vain.
Some use cases are described in Section 4.3.
iOAM can designate the flow to add the iOAM header and collect data on the flow forwarding path. The flow can have arbitrary granularity. However, processing the data can be a heavy burden for the network nodes, especially when some data needs to be calculated by the node (e.g., the transit delay). If the flow traffic is heavy, the node may not be able to handle the iOAM processing so many performance issues may occur, such as long latency and packet drop.
Although it is good for the OAM applications to gain the detailed information on every packet at every node, in many cases, such information is often repetitive and redundant. The large quantity of data would also burden the management plane which needs to collect and stream the data for analytics. It is also possible that some nodes cannot provide the requested data at all or are unwilling to provide some data for security or privacy concerns. So a trade-off is needed to balance the performance impact and the data availability and completeness.
To minimize the network impact, a network operator decides to collect the iOAM data only for initial and last flow packets (e.g., TCP packets with SYN, FIN, and RST flags).
A head node alternates two iOAM headers with each requesting a subset of iOAM data. Hence, each node on the flow path only needs to handle partial data. The requests can be balanced without exhausting the network nodes.
A node is temporarily under heavy traffic load. It is in danger of dropping packets if it tries to satisfy all the iOAM data requests. In this case, it would rather deny some requests than drop user traffic.
More examples are listed in Section 5.2.
Based on the observation in Section 2.1, we propose a method for data type encoding which can solve the current limitation and address future data requirements.
Bitmap is simple and efficient data structure for high performance data plane implementation. The base bitmap size is kept to be 16 bits. We use one bit to indicate a single type of data in a single format. The last bit in the bitmap (i.e., bit 15), if set, is used to indicate the presence of the next data type bitmap, which is 32 bits long. In the second bitmap, bit 31 is again reserved to indicate a third bitmap, and so on. With each extra bitmap, 31 more data types can be defined.
Figure 1 shows an example of the in-situ OAM header format with two extended OAM trace type fields. Except the OAM Trace Type fields, all other fields remain the same as defined in [I-D.brockners-inband-oam-data].
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Base OAM Trace Type |1| Length Field | Flags |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Extended OAM Trace Type 1 |1|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Extended OAM Trace Type 2 |0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Node Data List [] |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 1: Extended OAM Trace Type Header Format
The specification of the Base OAM Trace Type is the same as the OAM Trace Type in [I-D.brockners-inband-oam-data] except the last bit, which is defined as follows:
The OAM trace type fields are labeled as Base OAM Trace Type, Extended OAM Trace Type 1, Extended OAM Trace Type 2, and so on. The Base OAM Trace Type is always present. If no data type is asked by the application in Extended OAM Trace Type n and beyond, then the last bit in the previous bitmap is set to 1 and these extended fields are not included in the header. On the other hand, to eliminate ambiguity, if any data is asked for by the application in Extended OAM Trace Type n, then Extended OAM Trace Type 1 to (n-1) must be included in the header, even though no data type in these bitmaps are needed (i.e., all zero bitmap except the last bit).
The actual data in a node is packed together in the same order as listed in the OAM Trace Type bitmap. Each node is padded to be the multiple of 4 bytes.
New types of data can be potentially added and standardized, which demand new bits allocated in the OAM Trace Type bitmaps. Some examples are listed here.
The length of each data must be the multiple of 2 bytes. However, allowing different data type to have different length, while efficient in storage, makes data alignment and packing difficult.
If we can define the maximum number of data types that can be carried per packet, the offset of each data in the node can be pre-calculated and carried in the iOAM header. The overhead can be justified by the overall space saving of the node data list. Otherwise, each data's offset in the node must be calculated in each device, with the help of a table which stores the size of each data type. We can also arrange the bitmap to reflect the data availability order in the system (e.g., the bit for egress_if_id must be after the bit for ingress_if_id), so in a pipeline-based system, the required data can be packed one after one.
Bitmap is simple and support parallel processing in hardware, however, it is not the only option to support data type extension. For example, cascaded TLV can be used to support arbitrary number of new data types.
Based on the observation in Section 2.2, we propose a method to limit the size of the node data list.
A hop is a node on a flow's forwarding path which is capable of processing iOAM data. A segment is a fixed number hops on a flow's forwarding path. While working in the "per hop" mode, the segment size (SSize) and the remaining hops (RHop), is added to the iOAM header at the edge. Initially, RHop is equal to SSize. At each hop, if RH is not zero, the node data is added to the node data list at the corresponding location and then RH is decremented by 1. If RH is equal to 0 when receiving the packet, the node needs to remove (in incremental trace option) or clear (in pre-allocated trace option) the iOAM node data list and reset RHop to SSize. Then the node will add its data to the node data list as if it is the edge node.
Figure 2 shows the proposed in-situ OAM header format. The last bit (bit 31) in the Flags field is used to indicate the current header is a segment iOAM header. In this context, the third byte of the first word is partitioned into two 4-bit piece. The first piece is used to save the segment size and the second piece is used to save the remaining hops. This limits the maximum segment size to 15.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Base OAM Trace Type |0| SSize | RHop | Flags |1|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Node Data List [] |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 2: Segment iOAM Header Format
At any hop when RHop is equal to 0, the node data list is copied from the iOAM header. The data can be encapsulated and reported to the controller or the edge node as configured. The encapsulation and report method is beyond the scope of this draft but should be comply with the method used by the iOAM edge node.
The actual size of the last segment may not be equal to SSize but this is not a problem.
Segment iOAM is necessary in the following example scenarios:
Based on the observation in Section 1.3, the source edge node should be able to define either the period or the probability to add the iOAM header to the selected flow packet. In this way, only a subset of the flow/sec packets would carry the OAM data, which not only reduces the overall iOAM data quantity but also reduces the processing work load of the network nodes.
It is possible that even an iOAM capable node will not add data to the node data list as requested. In some cases, a node can be too busy to handle the data request or some types of the requested data is not available. Therefore, we propose to add two bitmaps, a valid node bitmap and a valid data bit, to the iOAM specification.
The Node Valid Bitmap is inserted before the Node Data List as shown in Figure 3. Each bit in the bitmap corresponds to a hop on the packet's forwarding path. The bits are listed in the same order as the hop on the packet's forwarding path. The bitmap is cleared to all zero at first. If a hop can add data to the Node Data List, the corresponding bit in Node Valid Bitmap is set to 1. The bit location for a hop can be calculated from the length field (e.g, the bit index is equal to SSize-RHop).The valid node data items in the node data list is equal to the number of 1's in the Node Valid Bitmap.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Base OAM Trace Type |0| Length Field | Flags |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Valid Node Bitmap |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| Node Data List [] |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 3: Segment iOAM Header Format
For each node data in the node data list, a Valid Data Bitmap is added before the node data. The number of bits in the Valid Data Bitmap is equal to the number of 1's in the OAM Trace Type bitmaps (excluding the next trace type bitmap indicator bits). When the bit is set, the corresponding data is valid in the node; otherwise, the corresponding data is invalid so the management plane should ignore it after the data is collected.
The size of the bitmap can be padded to two or four bytes, which allow up to 16 or 32 types of data to be included in a node.
We give some examples to show the usefulness of in-situ OAM sampling and data validation features.
There is no extra security considerations beyond those have been identified by in-situ OAM protocol.
This memo includes no request to IANA.
We would like to thank Frank Brockners and Carlos Pignataro for helpful comments and suggestions.
The document is inspired by numerous discussions with James N. Guichard. He also provided significant comments and suggestions to help improve this document.
| [I-D.brockners-inband-oam-data] | Brockners, F., Bhandari, S., Pignataro, C., Gredler, H., Leddy, J., Youell, S., Mizrahi, T., Mozes, D., Lapukhov, P. and R. <>, "Data Formats for In-situ OAM", Internet-Draft draft-brockners-inband-oam-data-02, October 2016. |
| [I-D.brockners-inband-oam-requirements] | Brockners, F., Bhandari, S., Dara, S., Pignataro, C., Gredler, H., Leddy, J., Youell, S., Mozes, D., Mizrahi, T., <>, P. and r. remy@barefootnetworks.com, "Requirements for In-situ OAM", Internet-Draft draft-brockners-inband-oam-requirements-02, October 2016. |