- NANOG - lists.nanog.org

Re: While I'm at it...
by Peter Lothberg 10 Dec '96

10 Dec '96

> > While I'm on a roll of exploiting NANOG's collective expertise, perhaps > someone could email the IOS level (and perhaps card revision level) that > they have had the most luck with on a 75xx with RSP2 with AIP-DS3 and OC3 > cards. > > | Derek Elder http://www.accessus.net V.P., CIO | > | djelder(a)accessus.net accessU.S., Inc. 888-637-3638 Ext. 222 | > > > Luck has left for very long ago if you are using ATM -:) --Peter

1 0

error rate
by Jun Wu 10 Dec '96

10 Dec '96

I think I might have seen this question before, so please excuse me if this is not appropriate and just ignore it. Could someone let me know what would be an acceptable error rates on a leased line, especially international (transoceanic) links? This would break down to transmission error on the fiber/satelite, electronic error on end equipments (includes DACS, DSU/CSU, routers etc). If any references could be provided, that would be even better! Thanks! Jun -- Jun (John) Wu | Voice: (703)818-5431 Supervisor - Global IP Systems & Services | Fax: (703)818-5282 Global One Communications L.L.C. | Email: jun(a)gsl.net Reston, VA 22096 | URL: http://wolfox.gsl.net/jun

1 0

Re[2]: Exchanges that matter...
by rosas＠NetEdge.COM 09 Dec '96

09 Dec '96

Mathew, NetEdge would be very interested in learning more about your performance observations. Can we get more data? If you'd like we could go offline and offer some analysis and suggestions/solutions. Also, to answer your question, the performance is doubled from FDDI to ATM in the newer Edge platform just released. In fact, existing ATM Connect FDDI cards with a possible upgrade (if needed, depending on revision) can be inserted into the new Edge platform to yield this extra performance. Let me know how we can help. Rocky Rosas Director - Technical Support NetEdge Systems Support: 800 NET-ATM1 rosas(a)netedge.com ______________________________ Reply Separator _________________________________ Subject: Re: Exchanges that matter... Author: Matthew Petach <mpetach(a)netflight.com> at internet_mail Date: 12/9/96 4:08 PM > > On Sun, 8 Dec 1996, Nathan Stratton wrote: > > |} > Every NAP with a GIGAswitch/FDDI offers full duplex FDDI; the MAEs, > |} > Sprint, PAIX, and you. Buy a full-duplex-capable card, install it, and > |} > you get full duplex. You, the NAP operator, do nothing; the devices > |} > negotiate in and out of full duplex mode themselves. > |} > |} Ah, if I have a FDDI connection into MAE-East I cant do full duplex > |} unless they enable it, it now is disabled. > > Perhaps you should talk to them again. Steve Feldman (MFS) has said > multiple times that FFDT is enabled on the MFS Gigaswitches. At the last > NANOG, he suggested people try out the 'new' feature implemented by Cisco > and send him feedback. As Stephen Stuart (PAIX) pointed out you have to > disable the feature as it's enabled by default. *cough* *cough* Hate to jump into the middle of this wonderful debate, but has anyone here been able to push more than 60-70mbs through the fddi interfaces in the NetEdges anyhow? We've upgraded to an oc3c line to MAE-West, and discoverd that lo and behold, you don't get that much more out of the link than you did with a DS3 because the fddi cards top off not much farther up. Full duplex is a nice thought, and would win us another 5-10mbs, but since we are far and away a content provider (traffic ratio outbound to inbound is close to 10:1 at MAE-West), what we're concerned about is being able to deliver the content people are asking for. Opening up the back channel from shared to full FDDI is a, well, "nice" feature, but isn't nearly as attractive as getting better FDDI cards for the NetEdges. On the flip side, we've had the same NE's on our prem from MFS for about 2 years now; does MFS have any plan to cycle them out for the newer, faster models that are being released? Is it likely that we would see performance improvements by having a newer box/card combo at both ends of our link? Just some ideas to toss out before the crowd... Thanks for any feedback! > -jh- Matt Petach Network Engineer, InterNex (working from home)

1 0

Re: Exchanges that matter...
by Joe Rhett 09 Dec '96

09 Dec '96

> It is actually quite hard in practice to push a FDDI switch > into trouble, although theoretically it should be simple, modulo > the interesting effects that token-withholding can produce. Hm. Not sure what you mean. FDDI can certainly be over-utilized, that's not hard. On the other hard, FDDI doesn't fall down and die like Ethernet does when congested. > In practice, the limiting factor will tend to be the per-port > bandwidth rather than the aggregate bandwidth. One would have Bingo.

7 8

MAE clarifications: Full Duplex FDDI, Congestion
by feldman＠MFSDatanet.COM 09 Dec '96

09 Dec '96

There's been a flurry of information about the MAEs over the weekend, and I'd like to clear up a few points: Full Duplex FDDI The DEC Gigaswitch supports full duplex FDDI by default, and MFS has *not* disabled this feature. At least one provider has already plugged in a Full Duplex capable interface and seen it work without any intervention on our part. There's no additional charge for use of full duplex FDDI, and there's nothing anyone has to do to "sign up" for it. I would ask that you notify me so I can watch more closely for difficulties, however. Cisco's full duplex FDDI interface is a module for the VIP2 card, which I believe is only supported on the 75xx and 70xx/RSP (but not 70xx "classic") platforms. I don't know of any other router vendors supporting full duplex FDDI, but they don't always check with me first! MAE Congestion There have been problems at both MAE East and West recently with apparent packet loss on trunks between Gigaswitches. In both cases, the links themselves were not full, but the packets were being dropped on ingress to one of the Gigaswitches. The apparent reason is head-of-line blocking in that switch (Giga1 at MAE East, the Ames switch at MAE West) caused by one or more providers filling their FDDI access links. In both cases, we have been working with those providers to find ways to relieve pressure on their attachment ports, and have made some configuration changes to reduce the amount of packet drops. We are also working with DEC to find solutions for the head-of-line problems. Steve Feldman MFS

1 0

Re: Building exchanges that matter ..
by Vadim Antonov 09 Dec '96

09 Dec '96

Joe Rhett <joe(a)Navigist.Com> wrote: [Gb ethernet] >...and any technology which experiences performance problems at 40% >utilization will is questionable at best... Last time i used switched ethernet it didn't experience any problems at 100% load. But you just can't do ATM w/o a switch. Talk about flexibility. --vadim

2 1

Re: Building exchanges that matter ..
by Joe Rhett 09 Dec '96

09 Dec '96

> Very fast ethernets are designed with two modes, standard > a standard CSMA/CD bus and a full-duplex collision-free mode. [ .. your lesson on basic networking deleted .. ] > ATM since one is stuck playing with forwarding based on > only one address space. Yah, I'm familiar with all this. I'm sitting behind one of those "Bad Notworks" boxes right now, actually. > I'm not sure what flexibility you're referring to wrt ATM > vs very fast ethernets, however the latter in combination > with software in one particular modern router that isn't > completely rocket-science can spoof rate-limited VCs > based on MAC addresses, which more-or-less duplicates > the one feature of ATM that I happen to like. Uh.. I'm not sure what you're asking, because I suspect the answer I'm about to give you'd have already considered, no? Ethernet -> Flat. Creating true architectural redundancy isn't possible without separate routes for each connection. Also, even full duplex Ethernet never achieves full bandwidth utilization due to the protocol implementation (hardware protocol). Advantage: All vendors support it, and is usually works without help. ATM -> Flat/Star/Web/whatever. You can create both bandwidth and redundancy without routing. Disadvantage: Every vendor supports it differently, and most of them only have 2 people that really know it. Never works without help. Also, congestion control between hetrogenous systems is a no-op.

3 2

Re: Exchanges that matter...
by Rob Liebschutz 09 Dec '96

09 Dec '96

> On Sun, 8 Dec 1996, Joe Rhett wrote: > > > > > > It is actually quite hard in practice to push a FDDI switch > > > into trouble, although theoretically it should be simple, modulo > > > the interesting effects that token-withholding can produce. > > > > Hm. Not sure what you mean. FDDI can certainly be over-utilized, that's > > not hard. On the other hard, FDDI doesn't fall down and die like > > Ethernet does when congested. > > > > > In practice, the limiting factor will tend to be the per-port > > > bandwidth rather than the aggregate bandwidth. One would have > > At the Atlanta-NAP we offer full duplex FDDI, why not try to get MFS to do > it? Cisco now has a full duplex FDDI card, so you can do 200 Mbs into the > NAP. Talk to Steve Feldman from MFS. At the October Nanog meeting he mentioned that their Gigaswitches (not sure if this is was just for MAE West) were ready to support Full Duplex FDDI and he invited volunteers to try it. Unfortunately, the GRF does not yet have a full duplex FDDI interface available, but when I asked about it, Ascend told me that they had implemented a CDDI interface in 3 weeks time to gain the business of of one customer. Rob

2 1

Re: Exchanges that matter...
by Vadim Antonov 07 Dec '96

07 Dec '96

Joe Rhett <joe(a)Navigist.Com> wrote: >The issue that everyone seems to be overlooking is network design >flexibility. Even if ATM has higher overhead, you can only grow a FDDI >network so much. It really isn't all that hard to max out most FDDI >switches. Nah. The issue is that ATM is a rather expensive way to achieve nothing. And, if you never heard about it before: FDDI is not what people use to send data over backbone lines, no matter what trade mags show in their cute pictures. --vadim

1 0

Re: Exchanges that matter...
by Vadim Antonov 07 Dec '96

07 Dec '96

salo(a)msc.edu (Tim Salo) wrote: > From: Paul A Vixie <paul(a)vix.com> > [...] > ... and they aren't subject to ATM's cell tax ... >I am surprised, (well, maybe not), that you aren't concerned about >the excessive overhead present in FDDI networks... FDDI is not a WAN technology. Losing 20% of bandwidth of a 10 ft piece of fiber is one thing. Losing 20% of bandwidth of a $3M/yr circuit is quite different. ATM as a LAN is probably ok, if it can compete with Gb Ethernet price/performance-wise. --vadim

2 1