| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 2092.1 | some answers... | BACHUS::DEVOS | Manu Devos NSIS Brussels 856-7539 | Thu May 29 1997 18:52 | 62 |
| Hi Gea,
Let's go !
- is gated mandatory if you want to enbale ASEROUTING?
Yes, gated is mandatory in that case.
- from what I see in the man pages gated reads /etc/gated.conf, does
this mean that the information in /etc/routes is no longer read?
I think you are confused by two different & unrelated features.
1) ASEROUTING is a feature which allows that a service (via its IP alias)
is advertised on two (or more) networks althought this IP alias is only
present on one single network. So the clients connected to the "other"
networks are able to "find" the service via these "other" networks.
After all, the data and processes of this service are the only
important features needed by the clients. This feature is existing from
ASE version 1.0.
2) Another feature has been added from ASE version 1.3 and need a special
content in the file /etc/routes. This feature is called "ASE Primary
and Backup networks". Before that feature, the ASE members communicated
with each other via the "member" network i.e. the one associated with
the hostnames of the members. If that network was going down, a Network
partition was detected even if "other" networks were present and
available for ASE members daemon communication. Because of that
partition, ASE was frozen i.e. no manual failover or asemgr actions
were possible during the time the partition is present. Since version
1.3, the ASE members can now communicate via "backup" networks in case
of failure of the primary network. This is done via the path_status.awk
script in /var/ase/lib which receives from the ASE alert script a
"warning" that a "route" between the members has failed and this awk
script reacts simply by changing the static route via another network.
To be able to change a route, it is necessary that /etc/routes allows to
set the initial routes at the boot time. Thus /etc/routes in this
context does NOT need the routed daemon. (By the way, the routed
daemon is a daemon process which simply records in the route table all
the routes advertized on the network; it does NOT change any route
by itself).
- is the information in /etc/routed read even if routed is no started?
Yes, according to man routes(4), the /etc/routes file identifies static
routes that are automatically added to the network routing tables with the
/usr/sbin/route add command. The /sbin/init.d/route script contains the
/usr/sbin/route add command that is executed for each entry in the
/etc/routes file when the network is restarted on the system or the system
is rebooted.
- I know that ASEROUTING is not mandatory for the service relocation, but
my customer would like to use it to spead up his client access.
Do you mean "to speed up" ? You can speed up the access with
ASEROUTING because the same service can be reached by more than one
disk service, but I am confuse by your question because the title of
your note is "netwwork failover". Do you mean speed up the access or
the network failover?
Hope it helps, Manu.
|
| 2092.2 | more questions | MLNCSC::VOCI | | Mon Jun 02 1997 11:52 | 48 |
| Hello Manu,
Thanks for your reply. A few more questions.
The customer has 2 server ASE with 2 network interface each, all the clients
have only one interface, ASEROUTING is enable and there are no routers on the
network.
He has seen the following scenario:
+-----------------------+
| |
tu0 tu0
171.31.6.3 171.31.6.9 171.31.1.15
+--------+ +--------+ +------------+
|Server A| |Server B| |Disk Service|
+--------+ +--------+ +------------+
tu1 tu1
171.31.1.3 171.31.1.9
| |
+-----------+-----------+ Netmask 255.255.255.0
| Both network cards are
tu0 monitored by ASE.
171.31.1.6
+--------+
| Client |
+--------+
- The service is associated to tu1 (172.31.1) on the server.
tu0 fails.
Since tu1 in up and running nothing happens to the service.
The only problem is for the clients on network 172.31.1 that can't access
the service.
The file /etc/routes is setup correctly on both servers.
Is it correct if I tell the customer to modify the ni_status_awk and
redirect the out-going network traffic with 'route delete' and 'route add'
on the server that has the broken network card ?
- If the client is doing a telnet on the service and the network card fails,
all the new telnet work correctly, while the first telnet (issued while the
failure was happening) hungs.
Do I have to force the new routing path also in this case?
- With ASEROUTING enable is there some king of network balancing beetween
the interfaces? are requests from the clients services from both the network
interfaces or only from one?
Thanks again,
Gea
|
| 2092.3 | | BACHUS::DEVOS | Manu Devos NSIS Brussels 856-7539 | Mon Jun 02 1997 17:26 | 88 |
| <<< SMURF::USERA:[NOTES]ASE.NOTE;1 >>>
-< ase >-
================================================================================
Note 2092.2 routed or gated - this is the question 2 of 2
MLNCSC::VOCI 48 lines 2-JUN-1997 10:52
-< more questions >-
--------------------------------------------------------------------------------
Hello Gea,
+-----------------------+
| |
tu0 tu0
171.31.6.3 171.31.6.9 171.31.1.15
+--------+ +--------+ +------------+
|Server A| |Server B| |Disk Service|
+--------+ +--------+ +------------+
tu1 tu1
171.31.1.3 171.31.1.9
| |
+-----------+-----------+ Netmask 255.255.255.0
| Both network cards are
tu0 monitored by ASE.
171.31.1.6
+--------+
| Client |
+--------+
- The service is associated to tu1 (172.31.1) on the server.
tu0 fails.
Since tu1 in up and running nothing happens to the service.
The only problem is for the clients on network 172.31.1 that can't access
the service.
>>>> Which tu0 is failing? If you mean the tu0 of the client, then it
>>>> is normal that it can not work. If it is the tu0 of any of the
>>>> cluster systems, then your client should continue to run.
The file /etc/routes is setup correctly on both servers.
>>>> Give us the content, so we can check it.
Is it correct if I tell the customer to modify the ni_status_awk and
redirect the out-going network traffic with 'route delete' and 'route add'
on the server that has the broken network card ?
>>>> No, absolutely no! According to your drawing and your description,
>>>> your client has only one network connection, and there is no
>>>> external router between the two networks. The route add and route
>>>> delete that you saw in my network failover script should be used
>>>> in the context of my network topology where 4 network interfaces
>>>> were ALL connected to the same logical backbone with bridges.
>>>> Maybe, you should give us more details on your network topology
>>>> and also what you try to achieve.
- If the client is doing a telnet on the service and the network card fails,
all the new telnet work correctly, while the first telnet (issued while the
failure was happening) hungs.
>>>> Again which card is failing ? If a running telnet has network
>>>> problem, it should abort after the TCP timeout.
Do I have to force the new routing path also in this case?
>>>> What do you mean by "new routing path"? If you mean "route add"
>>>> and "route delete" the answer is "No".
- With ASEROUTING enable is there some king of network balancing beetween
the interfaces? are requests from the clients services from both the network
interfaces or only from one?
>>>> No, DECsafe is not intended to make network load balancing.
>>>> But, you can spread your cclients on the two networks, doing so a
>>>> kind of static (as opposed to dynamic) load balancing.
According to your qquestion, I suggest you read or re-read the network
chapter of the ASE admin guide. You are still merging two different
things (ASEROUTING and Network failover.)
Describe your network, show which network your clients are connected
to and explain us what you want to do, it will be easier for us to tell
you how to do (if possible :-)
Manu.
Gea
|
| 2092.4 | more infos | MLNCSC::VOCI | | Wed Jun 04 1997 09:20 | 30 |
| Here is the network diagram that the customer has sent me.
Client ws6ap (mounts service ssdc_dh)
|
|
Subnet A------------------------------- Ethernet
(Primary ASE Network) | x <-- This link breaks!!
| |
|srv1ap | socatrs, ssdc_dh (disk service) | |
| |
----------- -------------
|Member A1| | Member A2 |
----------- -------------
soc_dh = disk service |srv1as |socbtrs
| |
Subnet B---------------------------- Ethernet
(Backup ASE Network) |
|
Client ws6as (mount service soc_dh)
file /etc/routes
-host srv1ap srv1ap
-host srv1as srv1as
-host socatrs socatrs
-host socbtrs socbtrs
In this scenario the client ws6ap can't no longer access service ssdc_dh.
|
| 2092.5 | Ah, AH ! | BACHUS::DEVOS | Manu Devos NSIS Brussels 856-7539 | Wed Jun 04 1997 18:49 | 30 |
|
Hello Gea,
The things are becoming clearer now !
To allow the client ws6ap to access the service ssdc_dh when the Subnet-A
network interface of System-A is down, you need two things:
1) The primary network being broken between the two systems, you
need to set a "backup" network via Subnet-B. This implies that you
have the /etc/routes files shown in your reply in BOTH systems and
that you have run asemgr to set the backup network up. This step
is only needed for the inter-node ASE communication, not for your
service directly.
2) The ws6ap client can only access the Subnet-A, so the only way to
maintain the service accessability is to fail it over to the
SYSTEM-B. You need to modify the /var/ase/lib/ni_status.awk script
in the SYSTEM-A to decide a failover of ssdc_dh to SYSTEM-B when
the socatrs interface is down. In the SYSTEM-B, you should test the
srv1ap interface. To have ASE doing the detection of these
interfaces, you should use asemgr and select these interfaces as
"monitored". The gated daemon and ASEROUTING variable are NOT needed
for this feature.
Is it clear ?
Manu.
|