http://forums.whirlpool.net.au/forum-replies.cfm?t=10510 Why Is ADSL so Unreliable? ---- Their Nortel equipment is hanging by a thread. for the last few months, I worked for a serious german competitor of Nortel's, who supply the IP aggregation devices for most of the largest DSL networks in the world (Deutsche Telekom, Korea Telecom, etc). The design and peformance issues, and the way Telstra got sucked into purchasing the Nortel equipment is really really sad. They have: 1. Serious serious serious performance limitations, especially with the aggregation of thousands of PPP sessions. The Shasta wasn't designed to aggregate DSL subscribers - it was meant to be a business 'leased line' type platform with a fraction of the number of connections. They cannot pass traffic among more than about 2,000 PPPoE sessions with any long term stability. Also, all of the traffic which comes off a Line card must go to a central switch processor, via a data bus , get processed, go back on the bus, and back on the line card. This is a very inefficient design - one which other vendors have overcome. This seriously limits throughput on the box - the maximum it can pass in a Tunnel is about 20Mbit/s. When I say tunnel, i mean any technology like L2TP, a VLAN, an IPSec tunnel, etc. 2. The shasta has no support for MPLS (multi protocol label switching). MPLS is a 'new generation' of building networks which alot of the large carriers use now. Telstra even use it! (Fancy that). It is REALLY useful in the provision of VPN's. 3. The shasta has piss poor support for dynamic routing protocols. No customer in asia is using dynamic routing on the Shasta as far as it is known. They are best suited in a statically routed (and therefore less scalable) environment. Nortel actually claim one of their routing protocols is LDAP - shows how much they know about routing!! 4. The shasta does NOT support ATM rate shaping. This means that to connect 20 DSLAMS to the DSL network, the Shasta has to have 20 OC-3 ports - OR, the ATM network has to support cell buffering (which is very expensive to implement in atm switches). 5. The switching architecture is a centralised one. This can be likened to a unix machine. When the system is loaded lightly, everyone enjoys good performance. When the system is bogged down, performance turns to rubbish. The actions of the other customers can seriously harm the performance of your link. Other vendors get around this by using a distributed processing architecture - using ASICs on the line cards themselves. 6. The population of interfaces is odd. You can have a lot of interfaces, with terrible performance per interface, or you can have a few interfaces, with nice peformance :) Where's the logic in that? 7. Software upgrades / problems require the entire system to be rebooted. This gets back to the centralised architecture again - because everything is controlled by a cnetral switch processor. Other vendors have the ability to reboot certain aspects (even certain cards) in the system independently of the entire box. 8. The shasta has zero support for IP Multicast. 9. A system with 16 STM1 ports (that is 16 x 155Mb/s ATM interfaces) requires 1850W of power. That is a hell of alot! (As a comparison, a certain vendor's box draws 1kW for 48 STM1 ports). 10. There is absolutely no command line interface to manage / configure the Shasta. It has a graphical management system. This makes provisioning veryyyy difficult on a large scale (ie no batch scripts) If the GUI system fails, both the management system AND the shasta must go down for a reload. 11. Nortel do all their performance stats with a very small number of interfaces. Remember I said it performs well with only a small number of interfaces? For instance, a test was done with 8 STM-1 ports, and it performed brilliantly. But, who would buy a box with 8 ports only???. Load more in, and performance really suffers badly. That should give you the idea ;) User #907 1646 posts Tom This nutcase posts a lot