Re: Ethernet bonding mode 5 only using one Slave adapter.
On 10/11/2013 2:42 AM, Muhammad Yousuf Khan wrote:
> [Cut].....
> Are dual and quad port Intel NICs available in your country?
>
> Not very easily but yes, we can arrange. i personally have PCIe 4 Port
> intel NIC.
> so this can be arranged.
I recommend Intel NICs because they simply work, every time, full
bandwidth, full Linux kernel support, great feature set, etc. Very high
quality, long lasting. I had an Intel Pro/100 in service in an MX mail
server for over 10 years. Still works.
...
> just a very basic question i am into virtualization for few years on
> Debian box.
> i never host a VM on external box. i have more then 10 nodes and all VMs
> are hosted on local Mdadm RAID drives.
> Just to have an idea. if you like to suggest. how many VM can be hosted on
> 1G link. i know your next statement will be "it depends upon the
> utilization of your VM and decision would be made on IO stats basis"
Yes, it does depend on exactly that.
> but just asking in general how many general VMs can be hosted on 1G LAN
> that are more or less untouched throughout the day.
If idle? As many as you can fit in memory up to the hypervisor limit,
or virtual IP address limit, if there is a limit on these. It's
possible to create VMs that have no network stack at all. In that case
there is literally no limit WRT the shared GbE link.
> and my big big time confusion is backup the VM from Virtualization terminal.
> lets say for a while 2 VM are running on 1GB link and i am taking a backup
> of a VM from virtual server. as the server is connected to external storage
> on 1 GB link. first virtual server will bring all the virtual drive data
> from External box to local RAM via same 1GB link on which VMs are hosted.
> it does mean that when backup will start all other VMs has to suffer?
> so even if 1 VM is running and we are making/creating a backup then how can
> we avoid chocking the link or bottle neck.
Ok, so apparently I misunderstood previously. I was under the
impression that you had an NFS storage server box, a backup server box,
and many physically boxes on which you were running virtual machines.
I.e. 6 or more computers connected to a GbE switch.
If I understand correctly now, all of your VMs are on one PC, and there
is an NFS server somewhere on the network where you store the files. Is
there a switch between the PC with all of the VMs, and the NFS server?
If so...
There are a couple of ways to address this:
1. Add another GbE interface on the PC and dedicate it to NFS
traffic. You should be able to bind the NFS client to a specific
IP address. This will require setting up source based routing
so NFS traffic only uses the new interface. Without source based
routing Linux will always use the first bound adapter for all
outbound traffic. This dedicates the current NIC to everything
other than NFS traffic, so the VMs have 1Gb/s for non-NFS traffic,
and 1Gb/s for NFS traffic, 2Gb/s aggregate. This would be my
preferred method. It's low cost, just a NIC and a cable. But
you have a steep learning curve ahead WRT Linux routing. A bonus
is you'll learn a lot about Linux networking in the process.
2. Implement QOS features in the switch, if it has them, to limit
the amount of bandwidth used by NFS traffic. The problem with
this method is that most switches don't allow this on a per port
basis, but on a VLAN basis. Which means you'd be limiting NFS
bandwidth everywhere, network wide, not just to the VM PC.
...
> any howto document on DRBD and GFS2 on debian? as i am using debian and
> only debian in overall environment.
> DRBD+GFS2 has got a native support on Redhat (as GFS2 is owned by Redhat).
> i do not have the experience nor confidence on stability of the both.
> i will be glad if you share any specific one with Debian.
DRBD and GFS2 are both kernel modules. Their configuration on Debian
should be little different than on any Linux distro.
> i found this
> http://pve.proxmox.com/wiki/DRBD
http://www.drbd.org/users-guide/ch-gfs.html
> the above is Primary/Primary installation means both drbd drives can be
> mounted. but there is a question.
> if i can mount in Primary/Primary mode on both the nodes then what is the
> need of GFS?
> just asking for my learning.
The key word here is "mount". Linux cannot mount a block device. DRBD
is a block device. Linux mounts filesystems. Filesystems reside on top
of block devices. No two hosts can mount the same filesystem on a
shared block device unless it is a cluster filesystem. Cluster
filesystem are designed specifically for this purpose. However, in the
real world, the block device under GFS2 and OCFS2 filesystems is most
often a LUN on a fiber channel or iSCSI SAN storage array, not DRBD.
...
> Thanks for sharing such a detail and very helpful email.
You're welcome.
--
Stan
Reply to: