 |
|||
     |
|||
|
Revision Notes for Router Software v8.01 rev 1
Date: 12/21/94
Subj: Revision Notes for 8.01 fix 1
============================================================================
SITE MANAGER COMPATIBILITY
============================================================================
Router version is managed by Site Manager version
v8.01 2.00
8.01 fix 1 2.00
============================================================================
The image for FRE hardware is: 1773483 Dec 20 12:40 boot.exe
The image for VME & AFN Floppy hardware is: 1651186 Dec 20 12:59 ace.out
The image for AN hardware is: 1544831 Dec 20 13:36 boot.exe
The image for IN & AFN Flash hardware is 1645144 Dec 20 14:07 boot.exe
----------------------------------------------------------------------------
Revision fix/8.01/1 corrects the following problems:
=============================================================================
CR 4270: Source Route Bridge.
In source routing environment Apple end stations issue test frame
with 0xAAAAF3 dsap_ssap_ctrl field to the router. The router doesn't
reply to this frame and Apple end station support doesn't come up
on the network.
CR 11754: IP.
When IP circuits are disabled/enabled, OSPF HELLO packets can be
sent to wrong subsystem causing unpredictable results (like
orphaned buffer log entries).
CR 11006: DLSw.
The loss of a number of segments in the IP cloud between 2 routers
running DLSw causes a high number (70+ %) of TCP retransmissions
for the rest of the life of the connection. Slots can also reset
with out of memory errors.
The out of memory resets were caused by the method of handling
broadcast frames which are not flow-controlled - if TCP was busy
DLS would try to save incoming broadcast traffic for as long as
there is memory on a slot. In some cases this would result in DLS
eventually using up all available local memory causing a slot to
reset. This problem has been resolved by the following improvements
to the configurability of DLS.
Three new MIB attributes were added to the wfDlsSlotEntry MIB object.
wfDlsCurrentMemory -- to reflect the amount of memory in kilobytes
currently used by DLS on a particular slot.
wfDlsMaxAllowedMemory -- this is a configurable parameter. It
reflects how much memory in kilobytes DLS
is allowed to use on a particular slot.
When this number is reached DLS would start
dropping incoming traffic. Thus preventing
a router from crashing. The default value
is -1, Which means infinity. (all available
memory).
wfDlsHiWaterMark -- to reflect the highest memory consumption
in kilobytes reached on a particular slot
by DLS since the last reboot.
The last attribute is intended to be used in the following way.
Since every customer environment is different and it is hard to
predict an average traffic load, the user should let a router run
with wfDlsMaxAllowedMemory set to the default value for sometime.
Assuming there were no unusual events during this period of time a
user should set wfDlsMaxAllowedMemory to wfDlsHiWaterMark or some
number close to it. For example: after booting wfDlsMaxAllowedMemory
is set to -1. wfDlsHiWaterMark is 0. wfDlsCurrentMemory is 0.
Looking at the above attributes after 3 days one sees
wfDlsMaxAllowedMemory is still set to -1. wfDlsCurrentMemory is 100
and wfDlsHiWaterMark is 300. Now if you set wfDlsMaxAllowedMemory
to 300, it would assure that in an emergency situation DLS would not
run a slot out of memory because of the unlimited consumption of
memory. The optimal amount of memory allowed to be taken by DLS
should be determined experimentally for the particular configuration.
Access to the above attributes is available from TI only for now.
The Site Manager support for this enhancement will be implemented
in a later release.
CR 10922: PPP.
AN is connected to a Network Systems router via sync port. Sync
port is configured for PPP with bridging and IP. IP comes up fine.
When the BNCP negotiation takes place the Network Systems router
sends the AN a BNCP configure reject for all mac types (FDDI
ETHERNET and TOKEN RING). When the AN receives the configure reject
a bus error Panic occurs.
CR 9839: Bridge.
Setting up Bridge Traffic Filters to drop IP broadcast traffic
while bridging and running IP 'Not Forwarding' mode does not
work. The Traffic Filters are not applied.
CR 9175: SMDS.
Router responds to SMDS hartbeat poll on logical link 0 instead
of logical link 1.
CR 10685: Source Route Bridge.
Running NetBIOS Caching may cause panic in sr_nb_process_are due
to RIF entries being aged while the same RIF entries are being
relearned.
CR 10684: Source Route Bridge.
Running NetBIOS Caching may cause a 'PANIC in mib_ent.c at line
710' due to RIF entries being aged while the same RIF entries
are being learned.
CR 10543: System.
Slot resets with 'System restart in tmo_wdog.c at line 51' when
XNS router learns many (500) routes after wan link goes down/up.
CR 10498: NML.
Slot running NML can bus error in tfind when bridging traffic .
CR 10412: Bridge.
When Learning bridge traffic is flooded over FDDI onto a Frame
Relay network the router at the remote side does not flood the
bridge traffic out its other interfaces which reside on the same
slot as the Frame Relay interface.
CR 10557: AppleTalk.
Slot routing AppleTalk can experience Tag Violation reset upon
receipt of illegal NBP tuples.
CR 9870: OSPF.
Dynamic disabling of an OSPF interface can cause a PANIC bus
error on non-backbone routers.
CR 9958: Vines.
Vines router does not respond to StreetTalk requests from clients
on serverless segments.
CR 10314: DECnet.
In particular configuration DECnet router sends incorrect level
2 routing message designating local area as 1 hop and some non
zero cost away.
CR 10790: System.
On an ASN the attribute 'HwModuleModRev' was 0 all the time
instead of displaying the true Net Module revision number.
CR 9690: OSPF.
The External Route Tag Field within an OSPF AS External Link
State Advertisement is inconsistant when the ASE is generated
from a 7 series router. The use of this field, although not
documented in the OSPF specification to this effect, in the 5
series is to list the IP address of the gateway from which an ASE
Type 2 was originally learned via RIP (or other protocol). The 5
series does this consistently and has proved to be an invaluable
troubleshooting tool when debugging a large OSPF network where
there are multiple routers advertising the same ASE Type 2s. The
7 series on the other hand, will use the field in the same way
initially, but seems to lose track of what data is actually being
plugged into the field if the ASE Type 2 is refreshed for any
reason. An example of when this might happen is if an ASE Type
2 is aged, flushed and then relearned and refreshed again. The
External Route Tag field may look fine in an advertisement prior
to and when it is actually flushed, but when a new instance is
generated, bogus data may appear in this field.
CR 10696: Bridge.
Learning bridge over token ring does not learn about station
moves as it should.
CR 10745: Switched Services.
Enhancement for Dial-on-Demand circuits to allow the configuration
of the inactivity timer to be in seconds instead of minutes. Note
that this change affects existing config files such that what the
network administrator previously configured in minutes is now a
value in seconds.
CR 9225: Bridge.
8743: Bridge traffic doesn't properly switch back to the primary line
from a dial backup one. The router may stop learning and taking
incoming traffic from the sync port. Also the router may loop
back broadcast packets in the sync port.
CR 10787: IPX.
IPX router does not age out rip entries received with a hop
count of 15.
CR 9999: IPX.
Slot can reset with 'Out of Memory' errors when IPX interface with
many (> 400) hosts entries goes down.
CR 10667: IPX.
IPX Routes May Fail to Age Properly. The Lost Route implementation
incorrectly forms the packet to be sent out in the case where that
route was learned to be dead (hop count of 16) from a different
interface than the one that the route was originally learned from.
CR 10819: XNS.
XNS can restart on the slot where one or both sync lines in a
multiline configuration go down/up.
CR 10746: IPX.
IPX router responds to Get Nearest Server request when sap
filters are configured on the same interface where both a
server and a client reside (10746).
CR 10859: IPX.
IPX router does not correctly forwarded IPX Netbios frames
(type 14h).
CR 10857: IPX.
IPX router doesn't update its SAP advertisement after socket
number changes.
CR 7221: IP.
IP and OSPF remain up when FDDI interface is disconnected.
CR 10616: IPX.
Accept option doesn't work properly for IPX Traffic Filters.
CR 10569: IP.
When deleting one IP address from a multinetted interface arp
will no longer function correctly. If an IP address gets deleted
from a multinetted interface all devices on remaining network
will be unreachable.
CR 10314: DECnet.
In particular configuration DECnet router sends incorrect level
2 routing message designating local area as 1 hop and some non
zero cost away.
CR 10029: Filters.
When configuring outbound filters based on source or destination
MAC address, the MAX address must be entered in canonical format.
The problem then arises that when entering a MAC address range in
canonical format, the high address range could end up with a lower
value than the low range and Site Manager will give a message
'Minimum Value Greater Than Maximum Value'. Added a new MIB variable
'wfCctOptsEntry.wfCctOptsMacFilterFormat'. The default is 1,
which is canonical. Setting this to 2 will allow entering MAC
filters in token ring format. (Note: All filters on the interface
must be in token ring format!). Site Manager support for this option
will be implemented in a future release.
CR 10724: Switched Services.
If multiple packets are sent to a dial on demand circuit in quick
sequence bus error/panic will occur before the modem dials out.
The packets in this case were ip encapsulated source route
explorer packets.
CR 10563: Vines.
When running Vines over Frame Relay with Vines Fragmentation
Protocol (FRP) disabled, Vines traffic will not be forwarded.
CR 10702: IP.
Although a Wellfleet router running version 7.80 and 8.0 software
can transport DHCP requests/replies, there are still areas where
Wellfleet routers are not 100% RFC1532 compliant.
1) The BootP Relay agent should only care about UDP BootP Server
packets (port 67). The current code registers and processes for
both BootPC and BootPS ports. A relay agent MUST silently discard
any received UDP message whose UDP destination port number is
BootPC (68) in paragraph 4.1 of RFC1532.
2) Seconds field should apply to request packets only so one can set
the seconds field to a non-zero number and still let 0 second
replies through. Our implementation checks both request and
replies. The Microsoft implementation has a minor problem of
not setting it to 0 initially and not incrementing it
subsequently (see paragraph 3.2 in RFC1532).
3) The newly defined Broadcast flag is not checked in the current
code, the correct behavior is to broadcast only if 'ciaddr'
field is zero and the flag is set. (See paragraphs 3.1.1 and
5.3 in RFC1532).
CR 10582: Switched Services.
The dial backup link is not initiated in the following cases:
1. When the modem is set to loopback.
2. When Rx wire is being disconnected either on the master or on
the slave side.
This has been resolved by enhancing the sync drivers to check
for the BOFL packets and detect the loopback condition. The
loopback is detected by comparing the source MAC address in the
BOFL packets with the sync interface's MAC address. When the
loopback is detected the sync line driver is brought down. When
it is trying to come back up it goes into loopback state and stays
in it until it receives the BOFL from the remote side.
A new MIB attribute 'wfSyncRemoteLpbkDetection' has been added to
the wfSyncEntry. The above actions will take place only if this
attribute is set to ENABLED. This has been done intentionally since
most customers are using loopback for verifying that the link
functions properly. Site Manager support for this option will
be implemented in release 2.10.
CR None: System.
System reliability is now improved on ACE32 processors by the
addition of a watchdog process which will restart a slot if a
single task runs for more than 16 seconds.
CR 10793: IPX.
IPX ping over token ring fails intermittently.
CR 10814: DECnet.
DECnet router does not generate triggered update when adjacency
comes up.
CR 10966: DECnet.
DECnet router may experience a hogs bufs condition which could lead
to multiple PANICS when a topology change occurs. The specific
topology change would be in the form of a number of adjacencies
going unreachable at the same time and the presence of at least one
other router sharing a common segment. What can happen is that when
the adjacencies time-out, DECNET, which does not support
split-horizon, counts to infinity to age out the routes. During
this time of counting to infinity hogs bufs can occur. Also,
counting to infinity takes more CPU cycles than it should.
CR 11231: IPX.
Multiple instances of the same IPX route will appear in the MIB
if a router has multiple connections to another router, even
with multi-path turned off.
CR 10860: Vines.
A network that routes Vines and also has end station support and
servers that are redundantly linked to the bridged network,
may show servers as unreachable on the remote slots where Vines
is routed.
CR 10885: DLSw.
DLSw may cause a MIB panic when deleting instances after
failure of multiple DLSw sessions.
CR 11369: IP.
IP router looses track (has no ARP entry and doesn't send ARP
request for new one) of host entries after another slot resets.
CR none: IPX.
When IPX responds to Get Nearest Server, in the case of a tie in
the hop count to a number of different servers, IPX router should
respond with the one that is first in alphabetical order, which
is not the case starting with 7.71 fix 7. This problem was
introduced by the fix for CR 10746.
CR None: System.
The message 'gate 0x0c1fe being killed by 0x00049 @ 0x304a2c3e
is dead - ignored' frequently appears in the log. Although this
message can be generated from many subsystems, one particular
problem involving HostCache which produced those messages is
now resolved.
CR 8101: System.
Slot Resets with 'checksum error found in old system log'. The
entire system log for that slot is then lost. System log validation
has been improved to be more discrimating when making its decision
to eliminate a slot's damaged log.
CR 11095: DLSw.
DLS, when configured for an ethernet interface erroneously
responds with a Disconnect Mode packet to sna bind requests
which are destined to hosts on the same local ethernet segment.
This problem has the effect of preventing sna sessions between
hosts on the local ethernet segment from working.
CR 11366: IPX.
IPX is incorrectly processing SAP names followed by NULL chars and
ending with garbage characters. The problem shows up as large SAP
tables with many duplicated entries which are all the same server.
This causes the router to send out many SAP packets with what
appears to be the duplicate server entries.
CR 11168: OSI.
OSI router fails to accept packets with pad bytes before the
NLPID when received over PPP interface.
CR None: IPX.
The maximum number of IPX SAP filters per interface that can be
configured has been increased from 50 to 150.
The Site Manager support for this enhancement will be implemented
in versions 2.00 fix 1 and 2.1.
Feedback | Service Search | ERC Copyright © Bay Networks,Inc., 1997. All rights reserved. |
/tr> |
