[f-nsp] NetIron MLX Experience..

[Richard A Steenbergen]

On Tue, Aug 08, 2006 at 10:33:00PM +0200, Gunther Stammwitz wrote:
> 
> I don't agree.
> 
> The MLX can handle up to 512K IPV4 Routes in hardware (=the FIB) and the
> total size of possible BGP4 IPV4-routes in the RIB is 2 millions. This means
> you can have up to four full views with 512k routes each.

Yes the MLX is a bit better off than the Cisco 3B's on CAM. Remember that 
512k IPv4 only counts if you don't use the resources anywhere else like 
v6, MPLS, etc. BTW all of the numbers for v4 RIB capacity, BGP peers, etc, 
are completely arbitrary and for marketing purposes only. The reality is 
based on the amount of ram in the system, the efficiency of he code being 
run on it, and how many stupid hard-coded assumptions have been made in 
said code that limits internal resources. The memory is just DDR333 PC2700 
too, $150 will get you a nice 2GB DIMM for the MGMT, same as what comes on 
the XMR.

> The foundry box is much more powerful than the Cisco gear you mentioned and
> cannot be compared since forwarding is done in hardware on the line cards
> and not via software like the 7600 does.

I think you're confused, nothing in the above statement is even close to 
true.

The architecture of these boxes are roughly the same, but there are 
advantages and disadvantages to each difference. The Foundry MLX/XMR uses 
a large number of centralized dumb cell-switching fabrics to interconnect 
cards, and distributes the routing intelligence to each linecard. The 
Cisco equiv on the other hand comes with centralied routing intelligence 
by default (PFC), and allows you to add daughter cards to distribute the 
routing (DFC) and achieve roughly the same performance numbers. The 
Foundry wins hands down in performance unless you stuff your 7600 cards 
with DFCs which add significantly to the cost, but the tradeoff is that 
you're buying loads of tcam you may not need, and you can't upgrade those 
cards or make them interoperate between MLX/XMR.

There is a tradeoff on density too. The Foundry wins hands down on density 
for 10GE by using half-width slots, but Cisco win on density for 1GE 
because of there is only enough physical space to put half of the 
interfaces the Foundry is capable of supporting onto the card.

Same thing in price. Bit for bit the Foundry is much cheaper when maxed 
out, but because it doesn't come in any "mid-range" capacity flavors you 
can end up spending 33-50% more in absolute dollars for an equiv 
Foundry solution because you're buying capacity you don't need.

Point being, for everything in life there is a tradeoff. The new Foundry 
platforms are actually very interesting boxes, worth taking a good hard 
look at by anyone, but at the end of the day boxes that move a bazillion 
bits under simple lab conditions are a dime a dozen. It is software that 
makes the product, this is something that folks like Cisco and Juniper 
understand but that most other vendors do not. That said, if your network 
is simple enough and you don't need all of those interesting features (and 
if you have to ask or hell if you are even reading this list at all you 
probably don't), the MLX/XMR may be the box for you.

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)