Backup Central

Hey Guys,

=20

I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.

=20

Our data center is setup like below.

=20

2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.

=20

My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.

=20

In switch1: tapelib=3Dx.x.158.x(main server NIC),
tapelib1=3Dx.x.200.x,tapelib2=3Dx.x.201.x

In switch2: tapelib3=3Dx.x.158.x, =
tapelib4=3Dx.x.200.x,tapelib5=3Dx.x.201.x

=20

On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.

=20

This should work if the "storage nodes" field is set to the same value - I
believe that if left as nsrserverhost then the actual backed up data will
go through the "primary" backup server NIC. We had a problem with the
networker oracle module that we couldn't get to talk to the secondary
interface but would send the backup data via the storage node field entry.
There should be no problems with 2 routers choosing the right path IF you
are using tapelib1=x.x.200.x and tapelib4=x.x.200.x where all the x's are
not the same.

On Fri, 7 Jan 2005 09:00:41 -0500, Joel Fisher <jfisher < at > WFUBMC.EDU> wrote:

Hey Guys,



I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.



Our data center is setup like below.



2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.



My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.



In switch1: tapelib=x.x.158.x(main server NIC),
tapelib1=x.x.200.x,tapelib2=x.x.201.x

In switch2: tapelib3=x.x.158.x, tapelib4=x.x.200.x,tapelib5=x.x.201.x



On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.



From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate. Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.



Thanks,



Joel Fisher




--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.

So solaris. Personally, I don't see why SunTrunking wouldn't work here,
and it would reduce some of the administration overhead you're having to
deal with.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

Without some extra work, Solaris does not support having multiple
interfaces on the same subnet (only one interface will be used for
outbound traffic, and no error detection or failover will take place).

So clients would send data to one interface, but all the traffic back to
it might come from the other interface on the subnet (possibly on the
other switch).

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.

My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.

Networker can...

In switch1: tapelib=x.x.158.x(main server NIC),
tapelib1=x.x.200.x,tapelib2=x.x.201.x

In switch2: tapelib3=x.x.158.x, tapelib4=x.x.200.x,tapelib5=x.x.201.x

On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.

From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate.

The routers/switches will direct inbound traffic to exactly one of the
server interfaces based on the IP address.

The server will send outbound traffic for a subnet out exactly one of
the interfaces (which one probably depends on configuration order).

Now since you're expecting most of the traffic to be the inbound data
(at least during a backup), this may be fine. Just understand you have
no failover, and proper functionality may depend on having both
interfaces configured.

You could use IPMP to get failover, but I'm not sure it'll help with
correlating the outbound data with the "correct" interface.

Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.

I've done this a lot, but only with one interface per subnet, or with
sun trunking.


Darren Dunham ddunham < at > taos.com
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Hello,
Two points:
1. 6 gigabit ethernet cards on a V880 server
contactyour Sun Account Manager for support of 6 GBE NICs on a V880.
There are retrictions on number of HBAs supported per server type.
Don't know it by heart... but there are some limitations.


2. Use nsrserverhost as storage node,
- Add all the server hostnames as aliases of the backup servers.
- Redirect your LAN based baced up data for file system backups using "server network interface" parameter.
- Additionnally, for modules based backups such as NMO, NMSAP, ..., update you RMAN script or you're SAP utl file to change the server name.

Regards,
Tarik EL MANSOURI

Howard Martin <howard.martin < at > EDS.COM> wrote:
This should work if the "storage nodes" field is set to the same value - I
believe that if left as nsrserverhost then the actual backed up data will
go through the "primary" backup server NIC. We had a problem with the
networker oracle module that we couldn't get to talk to the secondary
interface but would send the backup data via the storage node field entry.
There should be no problems with 2 routers choosing the right path IF you
are using tapelib1=x.x.200.x and tapelib4=x.x.200.x where all the x's are
not the same.

On Fri, 7 Jan 2005 09:00:41 -0500, Joel Fisher wrote:

Hey Guys,



I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.



Our data center is setup like below.



2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.



My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.



In switch1: tapelib=x.x.158.x(main server NIC),
tapelib1=x.x.200.x,tapelib2=x.x.201.x

In switch2: tapelib3=x.x.158.x, tapelib4=x.x.200.x,tapelib5=x.x.201.x



On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.



From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate. Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.



Thanks,



Joel Fisher




--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu


---------------------------------
Découvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos mails !
Créez votre Yahoo! Mail

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

I'm getting ready to upgrade my server and along with this
upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.

We upgraded those links instead, by using WDM on the dark fibre links
and normal etherchannel on the others. It seems easier and is probably
more effortless than trying to manually load balance and administer
different networks.

But I still haven't had the chance to disable the sun trunking software
in order to see how expensive it is CPU-wise.

//Oscar

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Tarik,
I believe that the "server network interface" only controls the Meta data
and that the back up data is controlled by "storage nodes".
With the NMO module (networker 6.1.3/4) the changes in scripts and util
files did not work - this was with a backup server that had 2 NIC's with
no route between them (not allowed to do routing on the servers).

On Fri, 7 Jan 2005 17:28:29 +0100, Tarik El Mansouri <etarik < at > YAHOO.COM>
wrote:

Hello,
Two points:
1. 6 gigabit ethernet cards on a V880 server
contactyour Sun Account Manager for support of 6 GBE NICs on a V880.
There are retrictions on number of HBAs supported per server type.
Don't know it by heart... but there are some limitations.


2. Use nsrserverhost as storage node,
- Add all the server hostnames as aliases of the backup servers.
- Redirect your LAN based baced up data for file system backups
using "server network interface" parameter.
- Additionnally, for modules based backups such as NMO, NMSAP, ...,
update you RMAN script or you're SAP utl file to change the server name.

Regards,
Tarik EL MANSOURI

Howard Martin <howard.martin < at > EDS.COM> wrote:
This should work if the "storage nodes" field is set to the same value - I
believe that if left as nsrserverhost then the actual backed up data will
go through the "primary" backup server NIC. We had a problem with the
networker oracle module that we couldn't get to talk to the secondary
interface but would send the backup data via the storage node field entry.
There should be no problems with 2 routers choosing the right path IF you
are using tapelib1=3Dx.x.200.x and tapelib4=3Dx.x.200.x where all the x's a=
re
not the same.

On Fri, 7 Jan 2005 09:00:41 -0500, Joel Fisher wrote:

Hey Guys,



I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.



Our data center is setup like below.



2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.



My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.



In switch1: tapelib=3Dx.x.158.x(main server NIC),
tapelib1=3Dx.x.200.x,tapelib2=3Dx.x.201.x

In switch2: tapelib3=3Dx.x.158.x, tapelib4=3Dx.x.200.x,tapelib5=3Dx.x.201.=
x



On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.



From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate. Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.



Thanks,



Joel Fisher




--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D=
*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D=
*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D


---------------------------------
D=E9couvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos
mails !
Cr=E9ez votre Yahoo! Mail

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D=
*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

Darren,
In my experience unless the router table is set to force traffic outbound
connections from a multiple NIC Sun will use a "least used" algorithm to
select between usable NICs this may be maintained for a particular session
but is not guaranteed to be the same the next time a connection is made,
and I believe that every Save set will use its own connection.

On Fri, 7 Jan 2005 08:20:15 -0800, Darren Dunham <ddunham < at > TAOS.COM> wrote:

I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.

So solaris. Personally, I don't see why SunTrunking wouldn't work here,
and it would reduce some of the administration overhead you're having to
deal with.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

Without some extra work, Solaris does not support having multiple
interfaces on the same subnet (only one interface will be used for
outbound traffic, and no error detection or failover will take place).

So clients would send data to one interface, but all the traffic back to
it might come from the other interface on the subnet (possibly on the
other switch).

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.

My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.

Networker can...

In switch1: tapelib=x.x.158.x(main server NIC),
tapelib1=x.x.200.x,tapelib2=x.x.201.x

In switch2: tapelib3=x.x.158.x, tapelib4=x.x.200.x,tapelib5=x.x.201.x

On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.

From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate.

The routers/switches will direct inbound traffic to exactly one of the
server interfaces based on the IP address.

The server will send outbound traffic for a subnet out exactly one of
the interfaces (which one probably depends on configuration order).

Now since you're expecting most of the traffic to be the inbound data
(at least during a backup), this may be fine. Just understand you have
no failover, and proper functionality may depend on having both
interfaces configured.

You could use IPMP to get failover, but I'm not sure it'll help with
correlating the outbound data with the "correct" interface.

Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.

I've done this a lot, but only with one interface per subnet, or with
sun trunking.


--
Darren Dunham ddunham < at > taos.com
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Darren,
In my experience unless the router table is set to force traffic outbound
connections from a multiple NIC Sun will use a "least used" algorithm to
select between usable NICs this may be maintained for a particular session
but is not guaranteed to be the same the next time a connection is made,
and I believe that every Save set will use its own connection.

Certainly if IPMP is used, then individual TCP connections will hash.

In the past, I've used many systems with multiple nics on the same
subnet without IPMP and usually one of them has almost no outbound
traffic. All the traffic was using the other interface.

This behavior may have changed with some release or patch, but I don't
have a box I can use to test it right now.

Darren Dunham ddunham < at > taos.com
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Lost track in the thread here.. would a Networker client connection get
confused under a simple setup (multiple independent NICS on the same LAN-
no trunking). I would think the outbound NIC server name (despite being
in the client's servers file) might confuse a Networker save session;
ie.. a reply comes back from different server name than it was initiated
on? Was that the point being made here?

Multiple NICs on the same LAN seems good from a central admin point of
view; one could manually load balance backup traffic based on storage
node name (or backup server interface name ) without having to move clients
to a different backup LAN.


Robert Maiello
Pioneer Data Systems

On Fri, 7 Jan 2005 09:51:48 -0800, Darren Dunham <ddunham < at > TAOS.COM> wrote:


Darren,
In my experience unless the router table is set to force traffic outbound
connections from a multiple NIC Sun will use a "least used" algorithm to
select between usable NICs this may be maintained for a particular session
but is not guaranteed to be the same the next time a connection is made,
and I believe that every Save set will use its own connection.

Certainly if IPMP is used, then individual TCP connections will hash.

In the past, I've used many systems with multiple nics on the same
subnet without IPMP and usually one of them has almost no outbound
traffic. All the traffic was using the other interface.

This behavior may have changed with some release or patch, but I don't
have a box I can use to test it right now.

--
Darren Dunham ddunham < at > taos.com
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

I use a multi-NIC master AND multi-NIC storage nodes. The top 3 things =
I can think of to assist in performance.
1. Add all master and storage node DNS values to all master and storage =
node /etc/hosts files.
2. Setup and use "nscd" (Name Server Caching Daemon) on all =
master/storage nodes to decrease DNS hits
3. On every client, set the "Server Network Interface" for =
Master/MetaData connections. Also on every client, set the "Storage =
Nodes" field by removing 'nsrserverhost' and putting in StorageNode =
identifier.
=09
NOTE: All my servers have an Enterprise (routed) network connection =
and a Backup (flat, non-routed) network connection(s). The Enterprise =
is used to telnet/GUI use/general interactions. The Backup network is =
used for MetaData and backup data transfers.

Example:
Master has 3 NICs
Master-A =3D=3D Enterprise connection
Master-B =3D=3D Backup Specific
Master-C =3D=3D Backup Specific

Storage Node One (SN1) has 2 NICs
SN1-N =3D=3D Enterprise connection
SN1-M =3D=3D Backup Specific

Storage Node Two (SN2) has 4 NICs
SN2-W =3D=3D Enterprise connection
SN2-X =3D=3D Backup Specific
SN2-Y =3D=3D Backup Specific
SN2-Z =3D=3D Backup Specific

All 3 servers would contain an /etc/hosts file that contains (at =
minimum):
x.x.x.x Master-A loghost
x.x.x.y Master-B
x.x.x.z Master-C
y.y.y.y SN1-N
y.y.y.z SN1-M
z.z.z.w SN2-W
z.z.z.x SN2-X
z.z.z.y SN2-Y
z.z.z.z SN2-Z

And when setting up client ABC to backup to SN2 -
In the ABC client config you would enter 'Master-B' as the "Server =
Network Interface" and you can enter all 3 SN2 values in the "Storage =
Nodes".
Storage Nodes:
SN2-X
SN2-Y
SN2-Z
- Or you could only add one and manually "balance" the load, like we =
have done.
- Or you can put "nsrserverhost" as the LAST entry, backup to master if =
the Storage Node ever goes down.

And if setting up client DEF to backup to SN1 -
Server Network Interface =3D Master-C
Storage Nodes =3D SN1-M

Probably much more rambling than I meant it to be. Hope this helps in =
some way.
--Ted


-----Original Message-----
From: Legato NetWorker discussion
[mailto:NETWORKER < at > LISTMAIL.TEMPLE.EDU]On Behalf Of Howard Martin
Sent: Friday, January 07, 2005 11:23 AM
To: NETWORKER < at > LISTMAIL.TEMPLE.EDU
Subject: Re: [Networker] Multiple NIC configuration


Tarik,
I believe that the "server network interface" only controls the Meta =
data
and that the back up data is controlled by "storage nodes".
With the NMO module (networker 6.1.3/4) the changes in scripts and util
files did not work - this was with a backup server that had 2 NIC's with
no route between them (not allowed to do routing on the servers).

On Fri, 7 Jan 2005 17:28:29 +0100, Tarik El Mansouri <etarik < at > YAHOO.COM>
wrote:

Hello,
Two points:
1. 6 gigabit ethernet cards on a V880 server
contactyour Sun Account Manager for support of 6 GBE NICs on a =
V880.
There are retrictions on number of HBAs supported per server type.
Don't know it by heart... but there are some limitations.


2. Use nsrserverhost as storage node,
- Add all the server hostnames as aliases of the backup servers.
- Redirect your LAN based baced up data for file system backups
using "server network interface" parameter.
- Additionnally, for modules based backups such as NMO, NMSAP, ...,
update you RMAN script or you're SAP utl file to change the server name.

Regards,
Tarik EL MANSOURI

Howard Martin <howard.martin < at > EDS.COM> wrote:
This should work if the "storage nodes" field is set to the same value =
- I
believe that if left as nsrserverhost then the actual backed up data =
will
go through the "primary" backup server NIC. We had a problem with the
networker oracle module that we couldn't get to talk to the secondary
interface but would send the backup data via the storage node field =
entry.
There should be no problems with 2 routers choosing the right path IF =
you
are using tapelib1=3Dx.x.200.x and tapelib4=3Dx.x.200.x where all the =
x's are
not the same.

On Fri, 7 Jan 2005 09:00:41 -0500, Joel Fisher wrote:

Hey Guys,



I'm getting ready to upgrade my server and along with this upgrade I =
had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.



Our data center is setup like below.



2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I =
had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to =
do
this.



My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.



In switch1: tapelib=3Dx.x.158.x(main server NIC),
tapelib1=3Dx.x.200.x,tapelib2=3Dx.x.201.x

In switch2: tapelib3=3Dx.x.158.x, =
tapelib4=3Dx.x.200.x,tapelib5=3Dx.x.201.x



On a client plugged into switch1 on the 200 vlan I would put =
"tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put =
"tapelib4"
in the "Server network interface" field.



From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate. Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.



Thanks,



Joel Fisher




--
Note: To sign off this list, send a "signoff networker" command via =
email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D=
*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

--
Note: To sign off this list, send a "signoff networker" command via =
email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D


---------------------------------
D=E9couvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour =
vos
mails !
Cr=E9ez votre Yahoo! Mail

--
Note: To sign off this list, send a "signoff networker" command via =
email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

Note: To sign off this list, send a "signoff networker" command via =
email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D=
*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu
=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D*=3D

Hello
If I understand well the customer environment decribed by Joel, there' no storage nodes involved.
The backup server have or will have several nics on several networks.
Clients_A in network_A will use backup server interface bck_A
Clients_B in network_B will use backup server interface bck_B
..
Then the setup is simple,
Add bck_A, bck_B, ... as aliases on bck (backup server client)
And for each client in Clients_A use storage nodes nsrserverhost and server network interface bck_A
....


How it works.
Backup sever will start the backup command through the nsrexecd daemons using the backup server routing table (netstat -rn) . bck will adress client, if client is on a Network_A and backup server has a NIC on network_A then it will use network A to access the client, otherwise it will use the default route or a specific route manually defined (route add host ....)

On the client side, the backup command will redirect its data to the server using the server network interface that simply replaces the server name.
As the server name can be hardcoded in RMAN scripts or SAP backups using backint (utl_file), the scripts and configuration files for those specific backups have to be updated.

Additionnaly for configuration with storage nodes with multiple network interfaces. The additionnal hotnames defined on those storage nodes need also to be defined as aliases of the storage node main clients. (Make sure that the backup server can access to the storage node using only one route otherwise you will meet issues with the "nsrmmd polling" and the backup server main daemon will keep restarting the nsrmmd storage node daemons). Then to redirect the data from the client to the storage node, use the desired storage node hostname in the "storage nodes" setting for the client. Here as if a specific server network interface is used, it will only be used to send the client metadata informations.



Best regards.
Tarik EL MANSOURI

"Reed, Ted G II [ITS]" <ted.reed < at > MAIL.SPRINT.COM> wrote:
I use a multi-NIC master AND multi-NIC storage nodes. The top 3 things I can think of to assist in performance.
1. Add all master and storage node DNS values to all master and storage node /etc/hosts files.
2. Setup and use "nscd" (Name Server Caching Daemon) on all master/storage nodes to decrease DNS hits
3. On every client, set the "Server Network Interface" for Master/MetaData connections. Also on every client, set the "Storage Nodes" field by removing 'nsrserverhost' and putting in StorageNode identifier.

NOTE: All my servers have an Enterprise (routed) network connection and a Backup (flat, non-routed) network connection(s). The Enterprise is used to telnet/GUI use/general interactions. The Backup network is used for MetaData and backup data transfers.

Example:
Master has 3 NICs
Master-A == Enterprise connection
Master-B == Backup Specific
Master-C == Backup Specific

Storage Node One (SN1) has 2 NICs
SN1-N == Enterprise connection
SN1-M == Backup Specific

Storage Node Two (SN2) has 4 NICs
SN2-W == Enterprise connection
SN2-X == Backup Specific
SN2-Y == Backup Specific
SN2-Z == Backup Specific

All 3 servers would contain an /etc/hosts file that contains (at minimum):
x.x.x.x Master-A loghost
x.x.x.y Master-B
x.x.x.z Master-C
y.y.y.y SN1-N
y.y.y.z SN1-M
z.z.z.w SN2-W
z.z.z.x SN2-X
z.z.z.y SN2-Y
z.z.z.z SN2-Z

And when setting up client ABC to backup to SN2 -
In the ABC client config you would enter 'Master-B' as the "Server Network Interface" and you can enter all 3 SN2 values in the "Storage Nodes".
Storage Nodes:
SN2-X
SN2-Y
SN2-Z
- Or you could only add one and manually "balance" the load, like we have done.
- Or you can put "nsrserverhost" as the LAST entry, backup to master if the Storage Node ever goes down.

And if setting up client DEF to backup to SN1 -
Server Network Interface = Master-C
Storage Nodes = SN1-M

Probably much more rambling than I meant it to be. Hope this helps in some way.
--Ted


-----Original Message-----
From: Legato NetWorker discussion
[mailto:NETWORKER < at > LISTMAIL.TEMPLE.EDU]On Behalf Of Howard Martin
Sent: Friday, January 07, 2005 11:23 AM
To: NETWORKER < at > LISTMAIL.TEMPLE.EDU
Subject: Re: [Networker] Multiple NIC configuration


Tarik,
I believe that the "server network interface" only controls the Meta data
and that the back up data is controlled by "storage nodes".
With the NMO module (networker 6.1.3/4) the changes in scripts and util
files did not work - this was with a backup server that had 2 NIC's with
no route between them (not allowed to do routing on the servers).

On Fri, 7 Jan 2005 17:28:29 +0100, Tarik El Mansouri
wrote:

Hello,
Two points:
1. 6 gigabit ethernet cards on a V880 server
contactyour Sun Account Manager for support of 6 GBE NICs on a V880.
There are retrictions on number of HBAs supported per server type.
Don't know it by heart... but there are some limitations.


2. Use nsrserverhost as storage node,
- Add all the server hostnames as aliases of the backup servers.
- Redirect your LAN based baced up data for file system backups
using "server network interface" parameter.
- Additionnally, for modules based backups such as NMO, NMSAP, ...,
update you RMAN script or you're SAP utl file to change the server name.

Regards,
Tarik EL MANSOURI

Howard Martin wrote:
This should work if the "storage nodes" field is set to the same value - I
believe that if left as nsrserverhost then the actual backed up data will
go through the "primary" backup server NIC. We had a problem with the
networker oracle module that we couldn't get to talk to the secondary
interface but would send the backup data via the storage node field entry.
There should be no problems with 2 routers choosing the right path IF you
are using tapelib1=x.x.200.x and tapelib4=x.x.200.x where all the x's are
not the same.

On Fri, 7 Jan 2005 09:00:41 -0500, Joel Fisher wrote:

Hey Guys,



I'm getting ready to upgrade my server and along with this upgrade I had
planned on implementing sun trunking and just having one big(4Gb) pipe
on the same IP that I use now. Well, after having a long discussion
with the networking guys here, because our core links are only 1Gb I
would be better of having the separate NICs on separate switches and
separate subnets.



Our data center is setup like below.



2 Cisco 6509 connected to 2 Cisco routers(not sure what model) via 1Gb
pipes.

Our servers mostly use 3 subnets x.x.158.x, x.x.200.x and x.x.201.x.

So in a perfect world I would get the best network performance if I had
6 NICs 1 on each subnet on each switch. That would keep the amount of
traffic that has to cross the shared 1Gb pipe way down.

I have a V880 8x1.2Mhz 16GB Memory, so I have the physical ability to do
this.



My question is: Is networker able to take advantage of a configuration
like this? I would set it up like below.



In switch1: tapelib=x.x.158.x(main server NIC),
tapelib1=x.x.200.x,tapelib2=x.x.201.x

In switch2: tapelib3=x.x.158.x, tapelib4=x.x.200.x,tapelib5=x.x.201.x



On a client plugged into switch1 on the 200 vlan I would put "tapelib1"
in the "Server network interface" field.

On a client plugged into switch2 on the 200 vlan I would put "tapelib4"
in the "Server network interface" field.



From what I've found in the archive, at least the initial connect will
also use the 158 NIC in switch1(no problem). My concern is that it'll
try to use both of the 200 NICs instead of the one I designate. Does
anyone have a similar setup that is functioning correctly? This would
complicate my backup configuration a bit, but I think the potential
performance gains are quite high.



Thanks,



Joel Fisher




--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu


---------------------------------
Découvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos
mails !
Créez votre Yahoo! Mail

--
Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu



---------------------------------
Découvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos mails !
Créez votre Yahoo! Mail

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu

Lost track in the thread here.. would a Networker client connection get
confused under a simple setup (multiple independent NICS on the same LAN-
no trunking). I would think the outbound NIC server name (despite being
in the client's servers file) might confuse a Networker save session;
ie.. a reply comes back from different server name than it was initiated
on? Was that the point being made here?

No. This is a TCP connection. As far as the client can tell, it's the
same server (source IP/source port). The fact that it is initially
delivered from the server via a different interface could increase the
path taken, but the client can't easily tell that this has occurred.

Multiple NICs on the same LAN seems good from a central admin point of
view; one could manually load balance backup traffic based on storage
node name (or backup server interface name ) without having to move clients
to a different backup LAN.

Sure. My only concern is if the server outbound traffic is not
distributed. During a restore, it could cause performance issues.

It also seemed like suntrunking was available originally. That should
give the same performance benefit as the multiple interfaces reducing
some of the maintenance concerns.


Darren Dunham ddunham < at > taos.com
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >

Note: To sign off this list, send a "signoff networker" command via email
should be sent to stan < at > temple.edu