MPTCP Working Group F. Duchene
Internet-Draft UCLouvain
Intended status: Experimental V. Olteanu
Expires: May 4, 2017 University Politehnica of Bucharest
O. Bonaventure
UCLouvain
C. Raiciu
University Politehnica of Bucharest
October 31, 2016
Multipath TCP Load Balancing
draft-duchene-mptcp-load-balancing-00
Abstract
In this document we propose several solutions to allow Multipath TCP
to better work behind load balancers.
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 May 4, 2017.
Copyright Notice
Copyright (c) 2016 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
Duchene, et al. Expires May 4, 2017 [Page 1]
Internet-Draft MPTCP LB October 2016
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Proposed solutions . . . . . . . . . . . . . . . . . . . . . 3
2.1. Per-server addresses . . . . . . . . . . . . . . . . . . 3
2.2. Embedding Extra Information in Packets . . . . . . . . . 5
2.2.1. Proposal 1 . . . . . . . . . . . . . . . . . . . . . 5
2.2.2. Proposal 2 . . . . . . . . . . . . . . . . . . . . . 6
3. Comparaison of the solutions . . . . . . . . . . . . . . . . 9
4. Recommandations . . . . . . . . . . . . . . . . . . . . . . . 9
5. IANA considerations . . . . . . . . . . . . . . . . . . . . . 9
6. Security considerations . . . . . . . . . . . . . . . . . . . 9
7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 10
8. References . . . . . . . . . . . . . . . . . . . . . . . . . 10
8.1. Normative References . . . . . . . . . . . . . . . . . . 10
8.2. Informative References . . . . . . . . . . . . . . . . . 10
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11
1. Introduction
Multipath TCP is an extension to TCP [RFC0793] that was specified in
[RFC6824]. Multipath TCP allows hosts to use multiple paths to send
and receive the data belonging to one connection. For this, a
Multipath TCP connection is composed of several TCP connections that
are called subflows.
Many large web sites are served by servers that are behind a load
balancer. The load balancer receives the connection establishment
attempts and forwards them to the actual servers that serve the
requests. One issue for the end-to-end deployment of Multipath TCP
is its ability to be used on load-balancers. Different types of load
balancers are possible. We consider a simple but important load
balancer that does not maintain any per-flow state. This load
balancer is illustrated in Figure 1. A stateless load balancer can
be implemented by hashing the five tuple (IP addresses and port
numbers) of each incoming packet and forwarding them to one of the
servers based on the hash value computed. With TCP, this load
balancer ensures that all the packets that belong to one TCP
connection are sent to the same server since each packet contains the
five-tuple used by the hash function.
Duchene, et al. Expires May 4, 2017 [Page 2]
Internet-Draft MPTCP LB October 2016
+--+---- S1
---|LB|---- S2
+--+---- S3
Figure 1: Stateless load balancer
With Multipath TCP, this approach cannot be used anymore when
subflows are created by the clients. Such subflows can contain any
five tuple and thus packets belonging to them will be load-balanced
to any server, not necessarily the one that was selected by the
hashing function for the initial subflow.
In this document, we propose several solutions to allow Multipath TCP
to work behind load balancers.
2. Proposed solutions
2.1. Per-server addresses
A first solution is to use two types of public addresses. The load
balancer uses a public address that is advertised in the DNS. This
address is used to establish the initial subflow of all Multipath TCP
connections. In addition to this address, a pool of addresses is
used for the servers behind the load balancer. One address of this
pool is assigned to each server behind the load balancer. This
server address is not announced in the DNS and only advertised by the
servers through the ADD_ADDR option.
The additional per-server address is used by the clients when they
wish to create additional subflows. Since each server has its own
public address, this ensures that the additional subflows are
directed to the corresponding server. For this solution, we need to
ensure that the client never use the public address of the load
balancer to initiate subflows. This can be achieved by a slight
modification to the MP_CAPABLE option described below.
To allow Multipath TCP to work for servers behind layer 4 load
balancers, we propose to use the reserved "B" flag in the MP_CAPABLE
option sent (shown in Figure 2 in the SYN+ACK. This flag informs the
other host that this address MUST NOT be used to create additional
subflows.
A host receiving an MP_CAPABLE with the "B" set to 1 MUST NOT try to
establish a subflow to the source address of the MP_CAPABLE. This
bit can also be used in the MP_CAPABLE option sent in the SYN by a
client that resides behind a NAT or firewall or does not accept
server-initiated subflows.
Duchene, et al. Expires May 4, 2017 [Page 3]
Internet-Draft MPTCP LB October 2016
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
+---------------+---------------+-------+-------+---------------+
| Kind | Length |Subtype|Version|A|B|C|D|E|F|G|H|
+---------------+---------------+-------+-------+---------------+
| Option Sender's Key (64 bits) |
| (if option Length > 4) |
| |
+---------------------------------------------------------------+
| Option Receiver's Key (64 bits) |
| (if option Length > 12) |
| |
+-------------------------------+-------------------------------+
| Data-Level Length (16 bits) | Checksum (16 bits, optional) |
+-------------------------------+-------------------------------+
Figure 2: Multipath Capable (MP_CAPABLE) Option
This bit can be used by servers behind a stateless load balancer.
The servers set the "B" flag in the MP_CAPABLE option that they
return and advertise their own address by using the ADD_ADDR option.
Upon reception of this option, the clients can create the additional
subflows towards these addresses. Compared with current stateless
load balancers, an advantage of this approach is that the packets
belonging to the additional subflows do not need to pass through the
load balancer.
To demonstrate the principle of an off path load balancer let's
consider a server behind a load balancer.
+-- net1 --+ +-- Load Balancer --+--- ADDR 1 ---+
| | | |
client --+ +--+ +--- Server
| | | |
+-- net2 --+ +------------- ADDR 2 -------------+
Figure 3: A server with 2 addresse.
As shown in figure Figure 3, this server has 2 IP addresses: 1 behind
the load balancer and 1 directly connected to the Internet. The
client sends a SYN containing an MP_CAPABLE option, the server
answers with a SYN+ACK containing an MP_CAPABLE with the "B" flag set
to 1. Upon reception of the SYN+ACK, the client will know that it
cannot establish any more subflow towards IP address. The server
will then advertise it's secondary address with an ADD_ADDR. Once
the client has established at least one connection to the secondary
Duchene, et al. Expires May 4, 2017 [Page 4]
Internet-Draft MPTCP LB October 2016
IP address, the server could elect to close the primary subflow or to
put it in backup mode.
2.2. Embedding Extra Information in Packets
Under some circumstances, addressing the individial servers via their
individial IPs is not desirable or feasible. To work around this
issue, we propose two mutually-exclusive solutions. They rely to
varying degrees on getting the client to embed connection or server-
identifying information in the packets that it sends out. This extra
information can be used statelessly by the loadbalancers.
Both solutions require modifications only to the server stack and
work well with existing MPTCP clients.
2.2.1. Proposal 1
Our first proposal revolves around controlling the destination port
that the client uses in all subflows aside from the initial one. It
is possible for the server to advertise an additional port via the
ADD_ADDR option [RFC6824]. This informs the client that it can send
an MP_JOIN to this new port and initiate a new subflow.
To take advantage of this, each server is be assigned a unique 16-bit
ID, which must be different from the port on which the service is
being hosted (e.g. 80). As soon as a connection is initiated, the
server sends an ADD_ADDR to the client advertising a new port equal
to said ID.
Packets that arrive at the loadbalancer are treated as follows:
o Packets destined to the port that the service is being hosted on
will be forwarded to a server based on a hash of the 5-tuple.
o Packets destined to any other port are forwarded to the server
whose ID matches the destination port.
This approach has two drawbacks:
o The client will most likely also try to initiate subflows using
the server's original port. Because these subflows are
loadbalanced based on a hash of their 5-tuple, they will almost
certainly reach a different server and break. (Using REMOVE_ADDR
to prevent the creation of these subflows would entail the
destruction of the original subflow.) This issue can be solved by
the adoption of the protocol modifications outlined in
Section 2.1.
Duchene, et al. Expires May 4, 2017 [Page 5]
Internet-Draft MPTCP LB October 2016
o If the client is behind a firewall that restricts access to
certain destination ports, it might not succeed in establishing
any new subflows.
2.2.2. Proposal 2
Our second proposal is to loadbalance packets based on the server's
token.
The token's most significant 14 bits are treated as a hash value for
the connection. They are embedded in all outgoing TCP timestamps,
and subsequently echoed back by the client. Incoming packets that do
not contain timestamps (such as FINs) are dealt with via redirection
between the servers.
2.2.2.1. Connection Initiation
The client initiates an MPTCP connection by sending a SYN with the
MP_CAPABLE option. Under normal operation, the server then picks a
random 64-bit key for the connection, and uses it to compute its
token.
To forward the packet appropriately, the load balancer must know the
token before deciding what server to send it to. To accomplish this,
we move the key generation to the load balancer. The connection's
token can be computed based on the generated key.
The load balancer places the generated key, along with the IP address
of the server that would be responsible for the subflow under normal
5-tuple hashing (which we call the alternate server IP) in an IP
option and forwards the SYN to the server.
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
+---------------+---------------+---------------+---------------+
| Type = 96 | Length = 16 | Unused |
+---------------+---------------+---------------+---------------+
| |
+ Server Key +
| |
+---------------+---------------+---------------+---------------+
| Alternate Server IP |
+---------------+---------------+---------------+---------------+
Figure 4: IP Option Used for MP_CAPABLE packets
Duchene, et al. Expires May 4, 2017 [Page 6]
Internet-Draft MPTCP LB October 2016
The figure above depicts the IP option that is inserted into the
MP_CAPABLE packet before it is sent to the server. We have chosen an
IP option despite the fact that the data contained therein pertains
to the transport layer, because TCP option space is very limited. IP
option type 96 is currently classified as reserved [RFC0791].
Upon receipt of the packet, the server uses the key provided to
compute the token for the connection. If no connection with the same
token exists, the server uses the key provided. Otherwise, it takes
a brute-force approach and randomly generates multiple keys and
selects one that yields a token with the same 14 highest-order bits.
The use of the alternate server IP will be discussed in a later
section.
2.2.2.2. Handling MP_JOIN packets
Additional subflows are initiated by the client by sending MP_JOIN
packets. These packets contain the server's token.
Similarly to how MP_CAPABLE packets are treated, the load balancer
uses an IP option to inform the server about which other server would
be responsible for the subflow under normal 5-tuple hashing.
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
+---------------+---------------+---------------+---------------+
| Type = 97 | Length = 8 | Unused |
+---------------+---------------+---------------+---------------+
| Alternate Server IP |
+---------------+---------------+---------------+---------------+
Figure 5: IP Option Used for MP_JOIN packets
IP option type 97 is also classified as reserved [RFC0791].
2.2.2.3. Embedding the token in the timestamp
The TCP timestamp option [RFC7323] is present in most packets and is
comprised of two fields: the TSval, which is set by the packet's
sender, and TSecr, which contains a timestamp recently received from
the other end.
Taking advantage of the fact that timestamps set by the server are
echoed back by the client, the server shifts its timestamp clock left
by 14 bits, and embeds the 14 highest-order bits of the token into
the 14 lowest-order bits of the TSval. When a packet with the ACK
Duchene, et al. Expires May 4, 2017 [Page 7]
Internet-Draft MPTCP LB October 2016
flag set and with the TS option present arrives at the loadbalancer,
it is forwarded based on the 14 least significant bits of the TSecr
field.
2.2.2.3.1. Impact on PAWS
Timestamps supplied by the server are used by the client for
protection against wrapped sequence numbers (PAWS). Note that for
Multipath TCP, the utilisation of the 64 bits DSN already protects
against PAWS.
We assume that the server uses a timestamp clock frequency of 1 tick
per ms, which is the highest frequency recommended by [RFC7323]. The
recycling time of the timestamp clock's sign bit is required to be
greater than the Maximum Segment Lifetime of 255 seconds. Given that
the clock ticks once every ms in increments of 2 ^ 14, its recycling
time is roughly 262 s, which is within the bounds set by the
standard.
While the quickly-increasing timestamp is benign to active subflows,
PAWS will still cause segments to be dropped if the subflow in
question had been idle for a period longer than the clock's recycling
time. To solve this, the server periodically sends keepalive
messages during idle periods.
2.2.2.4. Redirecting packets without timestamps
Some packets (most notably FINs) do not contain timestamps or any
other connection-identifying information. As such, they are
forwarded to a server based on a hash of the 5-tuple.
As seen in Section 2.2.2.1 and Section 2.2.2.2, whenever a new
subflow is setup, the server responsible for it (A) also knows which
other server (B) would be hit by the packets in case 5-tuple hashing
is used.
A will use a simple peer-to-peer protocol to inform B to setup a
redirection rule for the 5-tuple in question. The redirection rule
will be deleted by B either at A's request, after the subflow has
finished, or after a timeout. We do not discuss the specifics of the
protocol in this document.
Redirection of a packet is performed using IP-in-IP encapsulation.
Duchene, et al. Expires May 4, 2017 [Page 8]
Internet-Draft MPTCP LB October 2016
3. Comparaison of the solutions
Per-server addresses:
o Requires individual public addresses for each of the servers,
making IPv6 almost mandatory.
o Requires modifications to the clients and servers stack.
o Is transparent and works with today's load balancers.
o Doesn't need any modification to the applications.
o Disclose the real IP address of the servers.
o Allows to put the load balancer off-path.
Extra Information in Packets:
o Doesn't require an individual public addresses for each of the
servers.
o Requires modifications to the load balancers servers stack.
o Could be broken by a firewall blocking certain destination ports
(proposal 1) or changing the value of the timestamps (proposal 2).
o Doesn't need any modification to the applications.
o Doesn't disclose the real IP address of the servers.
4. Recommandations
5. IANA considerations
This document proposes some modifications to the Multipath TCP
options defined in [RFC6824]. These modifications do not require any
specific action from IANA.
6. Security considerations
Security considerations will be discussed in the next version of this
draft.
Duchene, et al. Expires May 4, 2017 [Page 9]
Internet-Draft MPTCP LB October 2016
7. Conclusion
In this document, we have described and compared two solutions to
load balance MultiPath TCP connections. We showed that these two
solutions have advantages and drawbacks and cover different network
configurations. Future versions of this draft will include more
solutions like the Application Layer Authentication and discuss
security considerations.
8. References
8.1. Normative References
[RFC0791] Postel, J., "Internet Protocol", STD 5, RFC 791, DOI
10.17487/RFC0791, September 1981,
<http://www.rfc-editor.org/info/rfc791>.
[RFC6824] Ford, A., Raiciu, C., Handley, M., and O. Bonaventure,
"TCP Extensions for Multipath Operation with Multiple
Addresses", RFC 6824, DOI 10.17487/RFC6824, January 2013,
<http://www.rfc-editor.org/info/rfc6824>.
[RFC7323] Borman, D., Braden, B., Jacobson, V., and R.
Scheffenegger, Ed., "TCP Extensions for High Performance",
RFC 7323, DOI 10.17487/RFC7323, September 2014,
<http://www.rfc-editor.org/info/rfc7323>.
8.2. Informative References
[I-D.ietf-mptcp-rfc6824bis]
Ford, A., Raiciu, C., Handley, M., Bonaventure, O., and C.
Paasch, "TCP Extensions for Multipath Operation with
Multiple Addresses", draft-ietf-mptcp-rfc6824bis-07 (work
in progress), October 2016.
[RFC0793] Postel, J., "Transmission Control Protocol", STD 7, RFC
793, DOI 10.17487/RFC0793, September 1981,
<http://www.rfc-editor.org/info/rfc793>.
[RFC1323] Jacobson, V., Braden, R., and D. Borman, "TCP Extensions
for High Performance", RFC 1323, DOI 10.17487/RFC1323, May
1992, <http://www.rfc-editor.org/info/rfc1323>.
[RFC6182] Ford, A., Raiciu, C., Handley, M., Barre, S., and J.
Iyengar, "Architectural Guidelines for Multipath TCP
Development", RFC 6182, DOI 10.17487/RFC6182, March 2011,
<http://www.rfc-editor.org/info/rfc6182>.
Duchene, et al. Expires May 4, 2017 [Page 10]
Internet-Draft MPTCP LB October 2016
[RFC7430] Bagnulo, M., Paasch, C., Gont, F., Bonaventure, O., and C.
Raiciu, "Analysis of Residual Threats and Possible Fixes
for Multipath TCP (MPTCP)", RFC 7430, DOI 10.17487/
RFC7430, July 2015,
<http://www.rfc-editor.org/info/rfc7430>.
Authors' Addresses
Fabien Duchene
UCLouvain
Email: fabien.duchene@uclouvain.be
Vladimir Olteanu
University Politehnica of Bucharest
Email: vladimir.olteanu@cs.pub.ro
Olivier Bonaventure
UCLouvain
Email: Olivier.Bonaventure@uclouvain.be
Costin Raiciu
University Politehnica of Bucharest
Email: costin.raiciu@cs.pub.ro
Duchene, et al. Expires May 4, 2017 [Page 11]