rfc3074.txt

最新的RFC

Network Working Group                                            B. Volz
Request for Comments: 3074                                      Ericsson
Category: Standards Track                                      S. Gonczi
                                                   Network Engines, Inc.
                                                                T. Lemon
                                                  Internet Engines, Inc.
                                                              R. Stevens
                                                      Join Systems, Inc.
                                                           February 2001


                      DHC Load Balancing Algorithm

Status of this Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2001).  All Rights Reserved.

Abstract

   This document proposes a method of algorithmic load balancing.  It
   enables multiple, cooperating servers to decide which one should
   service a client, without exchanging any information beyond initial
   configuration.

   The server selection is based on the servers hashing client Media
   Access Control (MAC) addresses when multiple Dynamic Host
   Configuration Protocol (DHCP) servers are available to service DHCP
   clients.  The proposed technique provides for efficient server
   selection when multiple DHCP servers offer services on a network
   without requiring any changes to existing DHCP clients.  The same
   method is proposed to select the target server of a forwarding agent
   such as a Bootstrap Protocol (BOOTP) relay.

1.  Introduction

   This protocol was originally devised to support a specific load
   balancing optimization of the DHCP Failover Protocol [FAILOVR].  The
   authors later realized that it could be used to optimize the behavior
   of cooperating DHCP servers and the BOOTP relay agents that forward
   packets to them.  The proposal makes it possible to set up each



Volz, et al.                Standards Track                     [Page 1]

RFC 3074              DHC Load Balancing Algorithm         February 2001


   participating server to accept a preconfigured (approximate)
   percentage of the client load.  This is done using a deterministic
   hashing algorithm, that could easily be applied to other protocols
   having similar characteristics.

2. Terminology

   This section discusses both the generic requirements terminology
   common to many IETF protocol specifications, and also terminology
   introduced by this document.

2.1.  Requirements 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 [RFC 2119].

2.2. Load Balancing Terminology

   This document introduces the following terms:

   Service Delay, SD
      A load balancing parameter, allowing delayed service of a client
      by a server participating in the load-balancing scheme, instead of
      ignoring the client.

   Hash Bucket Assignments, HBA
      A configuration directive that assigns a set of hash bucket values
      to a server participating in the load-balancing scheme.

   Server ID, SID
      An identifier that can be used to designate one of the
      participating Servers.  In the context of DHCP, the SID is the IP
      address or DNS name of the server.

   Service Transaction, ST
      A set of client-server exchanges that lead to a server providing
      or denying some service to a client.  Example: the DISCOVER/OFFER/
      REQUEST/ACK message exchange between a DHCP server and client is a
      service transaction.

   Service Transaction ID, STID
      An attribute of the individual client requests used for load-
      balancing.







Volz, et al.                Standards Track                     [Page 2]

RFC 3074              DHC Load Balancing Algorithm         February 2001


3.  Background and External Requirements

   Because DHCP clients use UDP broadcasts to contact DHCP servers, a
   client DHCPDISCOVER message may be received by more than one server.
   All servers receiving such a broadcast may respond to the client,
   letting the client choose which server it will use.

   When a BOOTP relay agent is used, it typically forwards or
   rebroadcasts client broadcasts to all configured servers, so a
   similar inefficiency is present.

   The optimization described allows a server to be chosen for each such
   transaction by performing a "serve" / "do not serve" computation.  A
   forwarding agent can perform the same computation to choose a
   forwarding destination.

   In either case, the choice of server can be computed, without the
   participants having to negotiate who is to respond.

   The approach is probabilistic in nature, because it is nearly
   impossible to foresee which client will request service next.  For
   short periods of time, the actual percentage of clients served by a
   given server will likely deviate from the desired percentage.  As the
   number of requests grows, the actual percentage of the load being
   handled by each server will approximate the configured percentage.

4. Overview

   DHCP servers MUST use the Client Identifier option as the STID if it
   is present.  If no Client Identifier option is present, the hlen
   field of the DHCP packet MUST be used as the length of the data to be
   hashed, and the contents of the chaddr MUST be the data to be hashed.
   At most the first sixteen bytes of the Client Identifier or chaddr
   are used.

   The proposal maps the STID into a hash value using the function in
   section 6.  The resulting hash value can then be used to decide who
   should respond to the request, or who the forwarding target should
   be.

   The provided hash function generates hash values 0 to 255, and yields
   a fairly even hash bucket distribution for random STID-s, and also
   for STID sequences that have some pattern.  Resource allocation is
   accomplished by assigning a set of specific hash values to each
   participating server.

   A server will only service a request if the STID hash of the request
   matches one of its assigned hash values.



Volz, et al.                Standards Track                     [Page 3]

RFC 3074              DHC Load Balancing Algorithm         February 2001


   Any hash buckets not assigned to servers will result in some client
   ST-s being entirely ignored.  (In some scenarios, this may be a
   desirable outcome.)  STID-s need not be unique, but should have
   sufficient variety to distribute load to each server.

   HBA-s MAY be transmitted as messages, encapsulated in messages of
   some other protocol, e.g., e-mail, or DHCP Failover Protocol option.

   DHCP server implementations may optionally be configurable to handle
   a case where load balancing is being done but the server that is
   supposed to respond is not available, or is out of suitable
   addresses.

   DHCP server implementations that provide this capability SHOULD set
   the DS (Delayed Service) configuration parameter to the number of
   seconds to wait after the client's first request has been sent before
   responding to a client, where the hash would not normally permit the
   client to be served.

   A DHCP server providing this capability SHOULD use the value in the
   secs field of the client request if its value is not zero.  Because
   some clients may not correctly implement the secs field, a DHCP
   server MAY keep track of the first instance of a client transaction
   to which it would not normally respond.  If the server receives a
   request from a client that has the same transaction ID as a
   previously recorded request, and if the secs field in the second
   packet is zero, the DHCP server MAY use the elapsed time (seconds)
   between the first and subsequent client request, instead of the secs
   field.

5. Operation

5.1 Configuration

   The configuration step consists of assigning hash values to available
   servers.  This is accomplished by providing one or more Hash Bucket
   Assignments (HBA-s).  These may come from a configuration file, the
   Windows NT registry, EEPROM, etc.  Alternatively, the hash bucket
   values could be assigned using some agreed upon algorithm.  E.g.,
   "Every odd value is serviced by server A and every even value is
   serviced by server B".

5.2 HBA Intended for a Server

   When configuring one specific server, an HBA in the form of a simple
   bit map of 32 octet values SHOULD be used.





Volz, et al.                Standards Track                     [Page 4]

RFC 3074              DHC Load Balancing Algorithm         February 2001


   The first octet in the HBA bitmap represents HBA values 0-7, the next
   byte values 8-15, and so on, with the thirty-second octet
   representing values 248-255.  In each octet, the least significant
   bit in that octet represents the smallest HBA value in that octet.

   Each bit of the HBA is associated with one possible hash value.  If a
   bit is set in the map, it means the recipient server MUST service
   each client request, where the STID yields the corresponding hash
   value.

   For example, if a server is configured with an HBA of the following
   32 octets:

            FF FF FF FF FF FF 00 00 ( 0   - 63 )
            FF FF FF FF FF FF FF FF ( 64 - 127 )
            00 00 00 00 00 00 00 00 (128 - 191 )
            00 00 00 00 00 00 00 00 (192 - 255 )

   then it MUST service any client requests where the STID hashes into
   the bucket values of 0 through 47 and 64 through 127.

5.3 Delayed Service Parameter

   The Delayed Service parameter is optional.

   If the parameter is not configured, the HBA sets up a strict Serve/Do
   not serve policy.

   If the parameter is configured, the server that is not supposed to
   serve a specific request (based on the HBA and the STID hash), is
   allowed to respond, after S seconds have elapsed since the client
   first attempted to get service.  A server MAY use the secs field in
   the BOOTP header for determining the time since the client has been
   trying to get service, or it MAY track repeated requests some other
   way.

5.4 HBA Intended for a Forwarder

   When configuring a forwarding agent, (e.g., BOOTP relay) HBA-s
   consisting of pairs of Server-ID / Hash Bucket values MAY be used.

   Here, the Server ID (SID) designates the server responsible for the
   specified Hash Bucket.  The forwarding agent forwards each client
   request, where the STID yields the specified hash value, to the
   server designated by the SID.






Volz, et al.                Standards Track                     [Page 5]