[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: 2 Network Cards

On Fri, Feb 01, 2008 at 08:59:38PM -0800, Raquel wrote:
> On Fri, 1 Feb 2008 22:45:30 -0500
> "Douglas A. Tutty" <dtutty@porchlight.ca> wrote:
> > Sure.  It would make sense if the hardware could keep up with two
> > NICs going full-out, so it depends on the network speed
> > (10/100/1000), the speed and quality of the NICs, and the power of
> > the computer.  Either way, every packet has to go through the
> > kernel and the firewall code.  I don't know which is faster.  It
> > takes a monster of a box to keep a GB ethernet saturated, yet alone
> > 2.
> > 
> > Look at it this way:
> > 
> > Lets say we're dealing with a 100 MB/s ethernet.  Lets say that all
> > the boxes on the network are all capable of saturating their 100
> > MB/s ethernet.  If all the NICs all try to talk at once, as long as
> > they are talking in pairs, then the switch should handle it.  In
> > this case, having two NICs in your box makes sense because, being
> > two virtual boxes, it is conceivable that two different client
> > boxes will want to talk to the server box at full speed, as long as
> > the server box can keep 2 100 MB/s NICs well fed while doing the
> > serving.  Also, as long as the switch back-plane has the
> > throughput.  
> > 
> > Remember, a second NIC will mean twice the hardware to be
> > interrupting the CPU.  
> > 
> > I would rather spend the money on one good NIC than two cheaper
> > ones.  
> > 
> > Doug.
> I think that I understand what you're saying.  However, what's the
> difference?  If the machine is capable of handling 15 VirtualHosts
> with 1 nic and 1 IP number, why can it not handle 15 VirtualHosts with
> 2 nics and 2 IP numbers?  What am I not understanding?

broadcasts will cause an interrupt on both interfaces. Something to remember 
eth0 will answer arp requests for the ip address on eth1 (there are sysctl's to 
stop this)

you will need to check to make sure you packets are leaving via eth1 (if eth0 
came up first). been a while since I have done this but your routing table is 
going to look something like

a.b.c.137 dev eth0 link mtu 1500 advmss 1460
a.b.c.138 dev eth1 link mtu 1500 advmss 1460
a.b.c.0/24 dev eth0 proto kernel scope link src a.b.c.137 mtu 1500 advmss 1460 
(<- this depends which one come up first [i believe])

some when it comes time to route an outbound packet it will go via eth0 

plus you will probably have something like this

default via a.b.c.1 dev eth0 mtu 1500 advmss 1460 (so all non local ip packet 
will go via eth0)

and depending on arp replies, if eth0 replies on behalf of eth1, all packets 
destined for the ip address on eth1 will come in via eth0 (again this can be 
changed with a sysctl)


> Scenario #1:
> domainA and domainB are being hosted on 1 machine with 1 nic.  Out of
> need, different IP numbers are needed, so eth0 is being aliased,
> creating eth0:1.
> Scenario #2
> domainA and domainB are being hosted on 1 machine with 2 nic.  Out of
> need, different IP numbers are needed, so nic#0 is eth0 and nic#1 is
> eth1.
> The only difference I can see is that, in essence, Scenario #1 is
> possibly "throttling" what reaches the CPU by virtue of what can get
> through the nic.
> -- 
> Raquel
> ============================================================
> The arc of the moral universe is long, but it bends towards justice.
>   --Martin Luther King, Jr.
> -- 
> To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org 
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org

"If we don't stop extending our troops all around the world in nation-building missions, then we're going to have a serious problem coming down the road."

	- George W. Bush
Boston, MA
First Presidential Debate with Al Gore

Attachment: signature.asc
Description: Digital signature

Reply to: