Veritas-bu

Re: [Veritas-bu] How to use different network interfaces

2007-10-31 14:57:30
Subject: Re: [Veritas-bu] How to use different network interfaces
From: "Rosenkoetter, Gabriel" <Gabriel.Rosenkoetter AT radian DOT biz>
To: "Barber, Layne Mr CTR US DISA CDM2" <layne.barber.ctr AT csd.disa DOT mil>, Veritas-bu AT mailman.eng.auburn DOT edu
Date: Wed, 31 Oct 2007 14:33:59 -0400
Stating which version of HP-UX would be helpful. I'll assume 11iV2, since it's the example I have handy, for the discussion of aggregating network ports, below.
 
You cannot force NetBackup to communicate with clients over a specific interface without carefully laying out the network. When the media server (even if your master and media are the same host, it's logically a media server in this context) communicates with a client, it simply resolves the client name to an IP address and then opens a socket to that client, subject to standard Unix routing. If you have one set of clients in 10.0.1/24, one set in 10.0.2/24, and one set in 10.0.3/24, then you can place one of your media server's interfaces in each of those subnets and know that for clients in each subnet the interface in that subnet will be the longest match route and traffic will flow through that interface. Any clients not in one of those three subnets will communicate with the media server through whatever its default route is. If your network is that well balanced and that much in your control, this will work. I don't think I've ever known that to be true for any organization that could afford NetBackup.
 
The "specified network interface" setting you're referring to is actually "REQUIRED_INTERFACE", and it won't work for this. You can't trump the OS's routing protocol that way. (If I understand it correctly, R_I's purpose was to keep a multi-homed master or media server from ever opening *listening* sockets on a specific network interface. This is useful for certain clustered server applications, but it's meaningless on a client.)
 
A far better approach is to make use of 802.11ad link aggregation (http://en.wikipedia.org/wiki/IEEE_802.3 has a decent explanation, see also http://www.ieee802.org/3/ad/). The way to do this under HP-UX is through their Auto Port Aggregation product (http://h20392.www2.hp.com/portal/swdepot/displayProductInfo.do?productNumber=J4240AA), which comes at a separate cost above the base operating system (so you may have to go buy it). APA can do a lot more (link failover, various types of auto- and reactive configuration, various other things supported in the 802.3 extensions), but all you really want here is the aggregation.
 
Here's a sample configuration to aggregate lan4, lan5, lan6, and lan7 on a 7420 running 11iv2 into a single "4" Gb (scare quotes because you shouldn't expect to see that speed: there's still more overhead with APA than the usual TCP overhead... it's still faster, just not necessarily 4x faster) lan900:
 
root@[REDACTED]:/etc/rc.config.d
[15] egrep -v '^#|^$' hp_apaportconf
HP_APAPORT_INIT_ARGS="HP_APAPORT_GROUP_CAPABILITY HP_APAPORT_PRIORITY HP_APAPORT_CONFIG_MODE HP_APAPORT_KEY HP_APAPORT_SYSTEM_PRIORITY"
HP_APAPORT_INTERFACE_NAME[0]=lan5
HP_APAPORT_CONFIG_MODE[0]=MANUAL
HP_APAPORT_INTERFACE_NAME[1]=lan6
HP_APAPORT_CONFIG_MODE[1]=MANUAL
HP_APAPORT_INTERFACE_NAME[2]=lan7
HP_APAPORT_CONFIG_MODE[2]=MANUAL
HP_APAPORT_INTERFACE_NAME[3]=lan4
HP_APAPORT_CONFIG_MODE[3]=MANUAL
root@[REDACTED]:/etc/rc.config.d
[16] egrep -v '^#|^$' hp_apaconf
HP_APA_START_LA_PPA=900
HP_APA_DEFAULT_PORT_MODE=MANUAL
HP_APA_INIT_ARGS="HP_APA_LOAD_BALANCE_MODE HP_APA_GROUP_CAPABILITY HP_APA_HOT_STANDBY HP_APA_MANUAL_LA HP_APA_INIT HP_APA_KEY"
HP_APA_INTERFACE_NAME[0]=lan900
HP_APA_LOAD_BALANCE_MODE[0]=LB_IP
HP_APA_MANUAL_LA[0]="5,6,7,4"


(As my prompt suggests, the config files live in /etc/rc.config.d.)

--
gabriel rosenkoetter
Radian Group Inc, Unix/Linux/VMware Sysadmin / Backup & Recovery
gabriel.rosenkoetter AT radian DOT biz, 215 231 1556

 


From: Barber, Layne Mr CTR US DISA CDM2 [mailto:layne.barber.ctr AT csd.disa DOT mil]
Sent: Wednesday, October 31, 2007 8:27 AM
To: Veritas-bu AT mailman.eng.auburn DOT edu
Subject: [Veritas-bu] How to use different network interfaces

First off, NBU 6.0MP5, HP-UX.

 

I have a master/media server with an STK L180 library and an Scalar24 library. The L180 is used for “normal” backups. The Scalar24 (which has only 2 LTO-3 drives) was added as part of a special project to migrate several clients to a different location. Basically backup those clients, send the tapes to new location and restore the data.

 

In an effort to increase the data through-put, I added 2 additional GB network connections. So now I have the following interfaces:

 

Jac01.x.x.x  Gigabit (original interface for non migrating clients)

Jac01v1.x.x.x  Gigabit

Jac01v2.x.x.x  Gigabit

 

How can I best configure NBU to have as much data as possible stream to the 2 LTO-3 drives in order to speed up the backups? I know I could create an STU on each interface and set the policies to use them, but that would seem to just make the jobs wait as other jobs tie up the drives. What about using the “specified network interface” for the clients? If I do that, how can I verify that the data is actually moving on the new interfaces instead of the original?

 

Help?

 

 

Thank You,
 
Layne Barber
MCSE 2K 2K3, Master-CNE, CompTIA A+, Security+
Systems Analyst I
Unisys Corp.
DISA Montgomery

 

_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
<Prev in Thread] Current Thread [Next in Thread>