NeuStar Comments on NANC 301 CO

1
NeuStar Comments on NANC 301 CO

• Charles Green
• NeuStar, Inc.
• 8 March 2000

2
NeuStar Comments on NANC 301 CO

• Description
• NPAC Monitoring of SOA and LSMS Associations via
  NPAC TCP Level Heartbeat
• Implementation by NPAC
• Apply HP/UX patch PHNE_17376
• enables support for TCP/IP heartbeats
• Enable the transport layer keepalive feature on
  the OTS/RFC1006 stack
• otsadm -k

3
NeuStar Comments on NANC 301 CO

• Recommendation
• Service Providers enable TCP Keepalive feature
• Implementation by SP
• Depends on OS
• Typically option implemented by system
  administrator
• Possibly requires an OS patch

4
NeuStar Comments on NANC 301 CO

• Purpose
• Poll opposite end of idle connection to insure
  availability
• Reduce duration of half-open connections
• Reduce phantom associations
• Cause aborts earlier than current detection of
  failures
• Detects some failures, does not correct cause

5
NeuStar Comments on NANC 301 CO

• Tunables
• tcp_keepstart
• The number of seconds that a TCP connection can
  be idle (that is, no packets received) before
  keep-alive packets will be sent attempting to
  solicit a response. When a packet is received,
  keep-alive packets are not longer sent unless the
  connection is idle again for this period of
  time.
• HP/UX default 7200 (two hours)
• NANC 301 recommends 60 (one minute)

6
NeuStar Comments on NANC 301 CO

• Tunables
• tcp_keepfreq
• The interval in seconds at which keep-alive
  packets will be sent on a TCP connection once
  they have been started. The receipt of a packet
  will stop the sending of keep-alive packets.
• If both tcp_keepfreq and tcp_keepstop are being
  increased, then tcp_keepstop should be increased
  first. If both are being decreased, then
  tcp_keepfreq should be decreased first.
• HP/UX default 75
• NANC 301 recommends 60 (one minute)

7
NeuStar Comments on NANC 301 CO

• Tunables
• tcp_keepstop
• The number of seconds keep-alive packets will be
  sent on a TCP connection without the receipt of a
  packet after which the connection will be
  dropped.
• If both tcp_keepfreq and tcp_keepstop are being
  increased, then tcp_keepstop should be increased
  first. If both are being decreased, then
  tcp_keepfreq should be decreased first.
• HP/UX default 600 (ten minutes)
• NANC 301 recommends 60 (one minute)

8
NeuStar Comments on NANC 301 CO

• tcp_keepstop
• With default values (tcp_keepfreq 75 and
  tcp_keepstop 600)
• If TCP sends a total of 8 packets without
  receiving any response from the remote TCP (that
  is, after 10 minutes have passed), TCP drops the
  connection.
• Setting tcp_keepfreq equal to tcp_keepstop may
  result in abort if the first heartbeat response
  is not received.
• NANC-recommended values may not permit retry to
  successfully maintain link.
• NeuStar recommends tcp_keepstop be set to at
  least twice tcp_keepfreq to insure at least one
  retry.

9
NeuStar Comments on NANC 301 CO

• tcp_keepstart
• Keepalive packets from one side are seen as link
  activity from opposite side.
• Setting tcp_keepstart equal on both sides of link
  could result in duplicate traffic with no benefit
• SP implementation of keepalive feature will
  permit most timely response to link failure
• SP must associate with NPAC, not vice versa
• NeuStar recommends NPAC tcp_keepstart be set at
  least one minute larger than tcp_keepstart value
  used by SPs.
• Predictable order of network events.