Dell PowerEdge w/ Intel AFT / Broadcom BASP

Brian A. Seklecki lavalamp at spiritual-machines.org
Thu Oct 6 13:24:01 PDT 2005


For the record on this, Dell claims that AFT/ALB is entirely software 
based.


On Wed, 5 Oct 2005, Brian A. Seklecki wrote:

> All:
>
> This may be better for freebsd-cluster at freebsd.org, but that list is kind of 
> ghost town, and this question is more a standards-based:
>
> Does anyone deploy Dell Poweredge in a HA configuration utilizing these 
> features?
>
> http://www.intel.com/network/connectivity/resources/technologies/load_balancing.htm
> http://www1.us.dell.com/content/topics/global.aspx/power/en/ps1q03_bhutani?c=us&cs=555&l=en&s=biz
> http://www.broadcom.com/drivers/faq_drivers.php#55
>
> Do we know what underlying standards and protocols compose these 
> "technologies"? 802.3ad, Cisco FEC?
>
> Intel AFT claims to provide redundancy over a "team" of NICs.  ALB claims 
> link aggregation; but they don't specify if they're doing it in hardware or 
> sofware (see Below)
>
> Broadcom BASP claims the same, given different terminology and vendor.
>
> I'm looking for a "fault tolerant" configuration for a HA cluster.  "Load 
> balancing" and/or "link aggregation" is not required.  I need to be able to 
> "team" two NICs into one Virtual NIC.  Each NIC connects to two redundant 
> managed switches, on which the connecting switch ports exist in the same VLAN 
> (which is then ISL/802.1q trunked between them).  Essentially
> the same ethernet segment.
>
> I see ng_one2many(4), but the man page doesn't really state what standard 
> that uses.  It seems to be all in-kernel magic (LACP and 802.3.ad aren't 
> mentioned in the man page); will this meet the above requirements?
>
> There were some ng_one2many(4) patches a while back to add more intellegence, 
> (FEC/802.3ad heartbeat like control protocol)
>
> http://marc.theaimsgroup.com/?t=107695977400002&r=1&w=2
> ...but no mention of them ever being commited.
>
> I see ng_fec(4) also, but I don't think that Cisco Ethernet Channel can occur 
> between two switches and one server (correct me if I'm wrong).
>
> I question the Hardware v.s. Software issue on the Intel NICs becase the Dell 
> PowerEdges Severs that happen to have Intel NIC Chipsets using em(4) (many 
> have Broadcom), seem to automatically try to "team" NICs when they're 
> connected to unmanaged PowerConnect switches, breaking ng_one2many logic. 
> They constantly alternate MAC addresses between the primary ethernet, the 
> secondary ethernet, and a 3rd 1-byte-off Virtual MAC.
>
> This automatic attempt to team seems like a hardware feature.  If it was a 
> software feature, in theory it wouldn't try to team w/o being instructed to?
>
> On the other hand, *managed* Dell PowerConnect switches feature something 
> called "LAG", which the docs describe as 802.3ad / LACP.
>
> I haven't tried ng_one2many on non-Dell or Dell Managed switches to see if 
> the MAC address "bouncing" problem persists, but I'll try that today.
>
> So the big question:
>
> *) Is the Windows/Linux-only software for configuring "teams" of NICs,
>    described in the URLs below, designed to configure a hardware level
>    feature that might have more intellegent link failure detection than
>    ng_many2one? (I.e., other than just lost carrier, say, STP storm
>    detection or excessive packet error thresholds).  Or is it software?
>
> *) If it is a hardware feature, could our em(4) driver be adapted or
>    could it possibly be configured using OpenManage via the Intel
>    IPMI/DMI/SMI whatever?
>
> *) Can Cisco FEC or 802.3ad provide reundancy between two switches and
>    one server w/ two NICs?  Will NetGraph ever have a 802.3ad module?
>
> *) What combination of Switch and NIC related teaming / failover technology
>    are known to be compatible with FreeBSD ?
>
> TIA,
> ~BAS
>

l8*
 	-lava

x.25 - minix - bitnet - plan9 - 110 bps - ASR 33 - base8


More information about the freebsd-questions mailing list