Return to BSD News archive
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!news.kei.com!news.mathworks.com!newsfeed.internetmci.com!nntp-hub2.barrnet.net!news1.digital.com!vixie!nnrp!vixie
From: vixie@gw.home.vix.com (Paul A Vixie)
Newsgroups: comp.unix.bsd.bsdi.misc
Subject: Re: Round Robin DNS??
Date: 19 Jul 1995 13:11:23 GMT
Organization: Vixie Enterprises
Lines: 457
Message-ID: <VIXIE.95Jul19061123@gw.home.vix.com>
References: <3uai48$2bq@lace.Colorado.EDU> <3uect4$mf8@park.uvsc.edu>
NNTP-Posting-Host: gw.home.vix.com
In-reply-to: Terry Lambert's message of 17 Jul 1995 19:15:48 GMT
> I think eventaully this will be dealt with at the protocol level.
right. have a look at this and let me know what you think.
Network Working Group Arnt Gulbrandsen
INTERNET-DRAFT Troll Technologies
Updates: RFC1035, RFC1183 Paul Vixie
Vixie Enterprises
May 1995
A DNS RR for specifying the location of services
Abstract
This document describes a DNS RR which specifies the location of the
server(s) for a specific protocol and domain (like a more general
form of MX).
Status of this memo
This document is an Internet-Draft. Internet-Drafts are working doc-
uments of the Internet Engineering Task Force (IETF), its areas, and
its working groups. Note that other groups may also distribute work-
ing 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 mate-
rial or to cite them other than as ``work in progress.''
To learn the current status of any Internet-Draft, please check the
"1id-abstracts.txt" listing contained in the Internet-Drafts Shadow
Directories on ds.internic.net (US East Coast), nic.nordu.net
(Europe), ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific
Rim).
This draft has file name "draft-gulbrandsen-dns-rr-srvcs-01.txt" and
expires on November 20, 1995.
Overview and rationale
Currently, one must either know the exact address of a server to con-
tact it, or broadcast a question. This has led to e.g.
ftp.whatever.com aliases, the SMTP-specific MX RR, and using MAC-
level broadcasts to locate servers.
The SRV RR allows a client to ask for a specific service/protocol for
a specific domain (the word domain is used here in the strict RFC1034
sense), and get back the names of any available servers. This allows
domain adminstrators to use several servers for a single domain, to
move servers with little fuss, and to designate some servers as pri-
mary and others as backups.
Gulbrandsen and Vixie [Page 1]
Expires November 1995 DNS Service Pointer RR May 1995
The format of the SRV RR
Here is the format of the SRV RR:
service.protocol.name ttl class SRV priority weight port target
(There is an example near the end of the draft.)
Service
The symbolic name of the desired service, as defined in Assigned
Numbers or locally.
Some widely-used services, notably POP, don't have a single uni-
versal name. If Assigned Numbers names the service indicated,
that name is the only name which is legal for SRV lookups. Only
locally defined services may be named locally.
The Service is case insensitive (it has to be, it's part of the
DNS look-up key).
Protocol
The symbolic name of the desired protocol. TCP and UDP are at
present the most useful values for this field, though any name
defined by Assigned Numbers or locally may be used (as for Ser-
vice). Case insensitive.
Name
The domain this RR refers to. The SRV RR is unique in that the
name one searches for is not this name; the example near the
shows this clearly.
TTL
Standard DNS meaning.
Class
Standard DNS meaning.
Priority
As for MX, the priority of this target host. A client MUST
attempt to contact the target host with the lowest-numbered pri-
ority it can reach; target hosts with the same priority SHOULD
be tried in pseudorandom order. The range is 0-65535. Domain
adminstrators are urged to use Priority 0 for the primary
server(s), to make the RR easier to read for humans using dig or
similar tools.
Weight
Load balancing mechanism. When selecting a target host among
Gulbrandsen and Vixie [Page 2]
Expires November 1995 DNS Service Pointer RR May 1995
the those that have the same priority, the chance of trying this
one first SHOULD be proportional to its weight. The range of
this number is 1-65535. Domain adminstrators are urged to use
Weight 0 when there isn't any load balancing to do, to make the
RR easier to read for humans (less noisy).
Port
The port where on this server host of this service. The range
is 0-65535. This is often as specified in Assigned Numbers but
need not be.
Target
As for MX, the domain name of the server host. There MUST be
one or more A records for this name. Implementors are urged, but
not required, to return the A record(s) in the Additional Data
section. Name compression is to be used for this field.
Domain adminstrator advice
Asking everyone to update their telnet (for example) clients when the
first internet site adds a SRV RR for Telnet/TCP is futile (even if
desirable). Therefore SRV will have to coexist with old-style A
record lookups for a long time, and DNS administrators should try to
provide A records to support old clients:
- Where the services for a single domain are spread over several
hosts, it seems advisable to have a list of A RRs at the same
DNS node as the SRV RR, listing reasonable (if perhaps subopti-
mal) fallback hosts for Telnet, NNTP and other protocols likely
to be used with this name. Some programs only try the first
address they get back from e.g. gethostbyaddr(), and we don't
know how widespread this behaviour is.
- Where one service is provided by several hosts, one can either
provide A records for all the hosts (in which case the round-
robin mechanism, where available, will share the load equally)
or just for one (presumably the fastest).
- If a host is intended to provide a service only when the main
server(s) is/are down, it probably shouldn't be listed in A
records.
- Hosts that are referenced by backup A records must use the port
number specified in Assigned Numbers for the service.
Currently there's a practical limit of 512 bytes for DNS replies.
Until all resolvers can handle larger responses, domain adminstrators
are strongly advised to keep their SRV replies below 512 bytes.
Gulbrandsen and Vixie [Page 3]
Expires November 1995 DNS Service Pointer RR May 1995
All round numbers, wrote Dr. Johnson, are false, and these numbers
are very round: A reply packet has a 30-byte overhead plus the name
of the service ("telnet.tcp.asdf.com" for instance); each SRV RR adds
20 bytes plus the name of the target host; each NS RR in the NS sec-
tion is 15 bytes plus the name of the name server host; and finally
each A RR in the additional data section is 20 bytes or so, and there
are A's for each SRV and NS RR mentioned in the answer. This size
estimate is extremely crude, but shouldn't underestimate the actual
answer size by much. If an answer may be close to the limit, using
e.g. "dig" to look at the actual answer is a good idea.
The "Weight" field
Weight, the load balancing field, is not quite satisfactory, but the
actual load on typical servers changes much too quickly to be kept
around in DNS caches. It seems to the authors that offering adminis-
trators a way to say "this machine is three times as fast as that
one" is the best that can practically be done.
The only way the authors can see of getting a "better" load figure is
asking a separate server when the client selects a server and con-
tacts it. For short-lived services like SMTP an extra step in the
connection establishment seems too expensive, and for long-lived ser-
vices like telnet, the load figure may well be thrown off a minute
after the connection is established when someone else starts or fin-
ishes a heavy job.
The Port number
Currently, the translation from service name to port number happens
at the client, often using a file such as /etc/services.
Moving this information to the DNS makes it less necessary to update
these files on every single computer of the net every time a new ser-
vice is added, and makes it possible to move standard services out of
the "root-only" port range on unix.
Usage rules
A SRV-cognizant client SHOULD use this procedure to locate a list of
servers and connect to the preferred one:
Do a lookup for QNAME=service.protocol.target, QCLASS=IN,
QTYPE=SRV.
If the reply is NOERROR, ANCOUNT>0 and there is at least one SRV
RR which specifies the requested Service and Protocol in the
reply:
Gulbrandsen and Vixie [Page 4]
Expires November 1995 DNS Service Pointer RR May 1995
for all such RR's, build a list of (Priority, Weight, Tar-
get) tuples
Sort the list by priority (lowest number first)
Create a new empty list
For each distinct priority level
While there are still elements left at this priority
level
Select an element randomly, with probability
Weight, and move it to the tail of the new list
For each element in the new list
query the DNS for A RR's for the Target or use any
RR's found in the Additional Data secion of the ear-
lier SRV query.
for each A RR found, try to connect to the (protocol,
address, service).
else if the service desired is SMTP
skip to RFC974 (MX).
else
Do a lookup for QNAME=target, QCLASS=IN, QTYPE=A
for each A RR found, try to connect to the (protocol,
address, service)
Notes:
- Port numbers SHOULD NOT be used in place of the symbolic service
or protocol names (for the same reason why variant names cannot
be allowed: Applications would have to do two or more lookups).
- If a truncated response comes back from an SRV query, and the
Additional Data section has at least one complete RR in it, the
answer MUST be considered complete and the client resolver
SHOULD NOT retry the query using TCP, but use normal UDP queries
for A RR's missing from the Additional Data section.
- A client MAY NOT discard any of the answers returned. RFC 974
allows clients to e.g. try to connect to just the 5 first MXes
returned: Such behaviour is NOT legal with SRV lookups.
Gulbrandsen and Vixie [Page 5]
Expires November 1995 DNS Service Pointer RR May 1995
- If the Additional Data section doesn't contain A RR's for all
the SRV RR's, the client MUST look up the A RR(s). (This hap-
pens quite often when the A RR has shorter TTL than the SRV or
NS RR's.)
- SRV RRs with Protocol TCP and Service SMTP override MX RR's.
This allows firewalled organizations with several SMTP relays to
control the load distribution using the Weight field.
- Designers of new protocols are urged to specify that SRV lookups
be mandatory for those protocols.
- Client resolvers may treat Weight 0 as equal to 1.
Fictional example
This is (part of) the zone file for asdf.com, a still-unused domain:
$ORIGIN asdf.com.
@ SOA server.asdf.com. root.asdf.com. (
1995032001 3600 3600 604800 86400 )
NS server.asdf.com.
NS ns1.ip-provider.net.
NS ns2.ip-provider.net.
ftp.tcp SRV 0 0 21 server.asdf.com.
finger.tcp SRV 0 0 79 server.asdf.com.
; telnet - use old-slow-box or new-fast-box if either is
; available, make three quarters of the logins go to
; new-fast-box.
telnet.tcp SRV 0 1 23 old-slow-box.asdf.com.
SRV 0 3 23 new-fast-box.asdf.com.
; if neither old-slow-box or new-fast-box is up, switch to
; using the sysdmin's box and the server
SRV 1 0 23 sysadmins-box.asdf.com.
SRV 1 0 23 server.asdf.com.
; SMTP - mail goes to the server, and to the IP provider if
; the net is down
smtp.tcp SRV 0 0 25 server.asdf.com.
SRV 1 0 25 mailhost.ip-provider.net.
MX 0 server.asdf.com.
MX 1 mailhost.ip-provider.net.
; NNTP - use the IP providers's NNTP server
nntp.tcp SRV 0 0 119 nntphost.ip-provider.net.
; addresses
server A 172.30.79.10
old-slow-box A 172.30.79.11
sysadmins-box A 172.30.79.12
new-fast-box A 172.30.79.13
Gulbrandsen and Vixie [Page 6]
Expires November 1995 DNS Service Pointer RR May 1995
; backup A records - new-fast-box and old-slow-box are
; included, naturally, and server is too, but might go
; if the load got too bad
@ A 172.30.79.10
A 172.30.79.11
A 172.30.79.13
In this example, a telnet connection to "asdf.com." needs an SRV
lookup of "telnet.tcp.asdf.com." and possibly A lookups of "new-fast-
box.asdf.com." and/or the other hosts named. The size of the SRV
reply is approximately 365 bytes:
30 bytes general overhead
20 bytes for the query string, "telnet.tcp.asdf.com."
130 bytes for 4 SRV RR's, 20 bytes each plus the lengths of "new-
fast-box", "old-slow-box", "server" and "sysadmins-box" -
"asdf.com" in the query section is quoted here and doesn't need
to be counted again.
75 bytes for 3 NS RRs, 15 bytes each plus the lengths of "server",
"ns1.ip-provider.net." and "ns2" - again, "ip-provider.net." is
quoted and only needs to be counted once.
120 bytes for the 6 A RR's mentioned by the SRV and NS RR's.
Refererences
RFC 1794: T. Brisco, "DNS Support for Load Balancing", 04/20/1995.
RFC 1713: A. Romao, "Tools for DNS debugging", 11/03/1994.
RFC 1712: C. Farrell, M. Schulze, S. Pleitner, D. Baldoni, "DNS
Encoding of Geographical Location", 11/01/1994.
RFC 1706: B. Manning, R. Colella, "DNS NSAP Resource Records",
10/26/1994.
RFC 1700: J. Reynolds, J. Postel, "ASSIGNED NUMBERS", 10/20/1994.
RFC 1536: A. Kumar, J. Postel, C. Neuman, P. Danzig, S. Miller, "Com-
mon DNS Implementation Errors and Suggested Fixes.", 10/06/1993.
RFC 1183: R. Ullman, P. Mockapetris, L. Mamakos, C. Everhart, "New
DNS RR Definitions", 10/08/1990.
RFC 1101: P. Mockapetris, "DNS encoding of network names and other
types", 04/01/1989.
RFC 1035: P. Mockapetris, "Domain names - implementation and specifi-
cation", 11/01/1987.
Gulbrandsen and Vixie [Page 7]
Expires November 1995 DNS Service Pointer RR May 1995
RFC 1034: P. Mockapetris, "Domain names - concepts and facilities",
11/01/1987.
RFC 1033: M. Lottor, "Domain administrators operations guide",
11/01/1987.
RFC 1032: M. Stahl, "Domain administrators guide", 11/01/1987.
RFC 974: C. Partridge, "Mail routing and the domain system",
01/01/1986.
Security Considerations
The authors believes this RR to be perfectly safe - or rather, not to
cause any new security problems. We assume that as the DNS-security
people invent new features, DNS servers will return the relevant RRs
in the Additional Data section when answering an SRV query.
Authors' Addresses
Arnt Gulbrandsen
Troll Tech
Postboks 6133 Etterstad
N-0602 Oslo
Norway
Phone: +47 22646966
Mail: agulbra@troll.no
Paul Vixie
Vixie Enterprises
Star Route 159A
Woodside, CA 94062
Phone: (415) 747-0204
Mail: paul@vix.com
Gulbrandsen and Vixie [Page 8]
--
Paul Vixie
La Honda, CA "Illegitimi non carborundum."
<paul@vix.com>
pacbell!vixie!paul (dont let the bastards grind you down)