jason at lixfeld
Jan 8, 2005, 3:44 PM
Post #25 of 25
(1657 views)
Permalink
|
On Jan 8, 2005, at 12:27 PM, Clayton Zekelman wrote:
> At 06:56 PM 1/7/2005, you wrote:
>> You say you're at 2000 users @ 20% CPU. What do you (or would you)
>> expect to see from your 7301 in terms of simultaneous users assuming
>> your ATM interface wasn't a factor? ATM != fragmentation issues
>> which overhead you don't have to worry about so there's more CPU to
>> work with there...
>
>
> My guess is that the 7301 would realistically be able to handle 8000
> users at 80% CPU load, if it scales linearly, given our current mix of
> traffic. An ATM OC3c interface is capable of around 4500 sessions
> given our current traffic mix. This is with the box handling only LNS
> functions. If you do traffic shaping, packet filtering, or tunnel
> switching in a LAC/LNS combined application, I would suspect the CPU
> utilization would go up somewhat. How much isn't clear.
Interesting. That's just over 29kb/s per subscriber. I've heard from
another wholesaler that their traffic patters are in the area of 50kb/s
per subscriber. Are those resi numbers or corporate numbers?
>> Are there architectural differences between DSL in Canada and DSL in
>> the US? In Canada, if you're with Bell your option is PPPoE over
>> L2TP and that's it. I seem to remember however, that some time ago
>> in the US you had options such as straight PPPoE w/o L2TP and
>> straight L2TP w/o PPPoE. Services like these would allow you to fit
>> more subscribers on a box, hence justifying a 16k BBA license on a
>> 7301 where in Bell land up here in Canada, you wouldn't be able to
>> push 16k sessions through a 7301 so an 8k BBA license should suffice,
>> no?
>
> I honestly don't know what the various LEC's in the US do for DSL
> aggregation. I'm sure it varies. We use the Bell (former Nexxia)
> L2TP network over ATM. The problem with the current L2TP over
> Ethernet LAN Extension architecture is precisely what you pointed out
> earlier - fragmentation. A 7301 running in that situation would
> likely have less headroom on the CPU unless can eliminate sources of
> fragmentation by using Adjust-MSS or otherwise limit the size of
> packets traversing the L2TP tunnels - which themselves take up CPU
> power.
>
> An 8000 user BBA license for a 7301 would likely be sufficient. Even
> if I could get more traffic into the box using ATM (thus avoiding
> fragmentation issues), I don't like pushing the CPU up as high as 80%.
> I like having headroom to be able to withstand a DoS attack if
> necessary.
Which brings me back to my original comment about a 7301 vs a 7206 :)
> In our particular application however, the 7301 is probably a good fit
> - the extra GigE ports we have can be used to aggregate PPPoE traffic
> from our co-locates in the Bell CO's. We have dark fiber which would
> allow us to use "IP" based (really Ethernet) DSLAM's in the future
> (currently we use ATM DSLAMs to bring PPPoE sessions into our C6400's
> in areas where we have co-locates, rather than using Bell's L2TP
> service).
Wow, that's an interesting model. Isn't colocate access expensive (not
to mention the dark fiber) Bell doesn't cross connect to your DSLAM
right? so for every DSL turn up you have, you need to do a truck role?
Or do you bundle truck rolls?
>> On Jan 7, 2005, at 6:32 PM, Clayton Zekelman wrote:
>>
>>>
>>> Its pretty close. The place where the costs start to add up on a
>>> 7206 is when you start to add redundant DC power supplies. The 7301
>>> is a single DC supply, with dual inputs, vs. the 7206 which is a
>>> true redundant supply. I think the SmartNet is a bit more costly on
>>> the 7206 as well.
>>>
>>> You're definitely right, the 7206 is certainly a more flexible
>>> option when it comes to multiple interfaces. We considered this,
>>> but figured we would end up burning more CPU horsepower later as we
>>> started using the processing capacity of the device more (filtering,
>>> shaping, etc...), so the number of interfaces wouldn't matter as
>>> much.
>>>
>>> I'd rather keep the boxes lightly loaded, and the $4k or so saved by
>>> going with the 7301 was worth the tradeoff.
>>>
>>> At 06:16 PM 1/7/2005, you wrote:
>>>> Totally off the cuff here, and not necessarily pertaining to an LNS
>>>> (because an LNS in this configuration would be bad on so many
>>>> levels) but the NPE-G1 is spec'd for 16k connections (but I have no
>>>> idea what the real world shows the performance of this CPU to be).
>>>> AFAIR, the 7301 has the same CPU but is limited to the number of
>>>> interfaces you can drop into it. If you get say a 7206 with a G1,
>>>> it would be cheaper than a 7301, no? (or maybe slightly more
>>>> expensive) but you can put 4 OC3 cards in it (any more and you run
>>>> out of BW points).
>>>> At the end of the day, wouldn't that be cheaper than rack and
>>>> stacks of 7301s? (in terms of both cost and rack space?).
>>>>
>>>> On Jan 6, 2005, at 2:08 PM, Clayton Zekelman wrote:
>>>>
>>>>>
>>>>> We're using a pair of 7301's as LNS's.
>>>>>
>>>>> Users are coming in on ATM-OC3c (PA-A3-OC3) on L2TP tunnels.
>>>>>
>>>>> One of our boxes is running with 2000 users at around 20% CPU load.
>>>>> I'll probably run out of bandwidth on the ATM PA before I run out
>>>>> of CPU power. Alternatively, the box also has 3 Gig/FastE ports,
>>>>> so if your sessions come in that way, you won't run into the
>>>>> bandwidth limitation of a single PA.
>>>>>
>>>>> I'm cutting over the second 7301 on the weekend, replacing my
>>>>> Cisco 6400's.
>>>>>
>>>>> We're using 12.3(11)T IP IOS
>>>>>
>>>>> I'm quite happy with the 7301's so far.
>>>>>
>>>>> At 11:58 AM 1/6/2005, you wrote:
>>>>>> I'm getting confused here, and was hoping someone could shed some
>>>>>> light on this.
>>>>>>
>>>>>> I've been trying to spec up an LNS. I've two options on the
>>>>>> table,
>>>>>> a 7300 series which I'm told will support 8000 users, and a 7600
>>>>>> series
>>>>>> which I'm told will support 16k users. The price difference is
>>>>>> less
>>>>>> 10%.
>>>>>>
>>>>>> It would seem to make sense to use the 7600 - is it possible one
>>>>>> of the
>>>>>> resellers is missing something?
>>>>>>
>>>>>> Thomas
>>>>>>
>>>>>> _______________________________________________
>>>>>> cisco-bba mailing list
>>>>>> cisco-bba [at] puck
>>>>>> https://puck.nether.net/mailman/listinfo/cisco-bba
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> No virus found in this incoming message.
>>>>>> Checked by AVG Anti-Virus.
>>>>>> Version: 7.0.300 / Virus Database: 265.6.8 - Release Date:
>>>>>> 1/3/2005
>>>>>
>>>>> ---
>>>>> Clayton Zekelman
>>>>> Managed Network Systems Inc. (MNSi)
>>>>> 344-300 Tecumseh Rd. E.
>>>>> Windsor, Ontario
>>>>> N8X 5E8
>>>>>
>>>>> tel. 519-985-8410
>>>>> fax. 519-258-3009
>>>>>
>>>>>
>>>>> --
>>>>> No virus found in this outgoing message.
>>>>> Checked by AVG Anti-Virus.
>>>>> Version: 7.0.300 / Virus Database: 265.6.8 - Release Date: 1/3/2005
>>>>>
>>>>> _______________________________________________
>>>>> cisco-bba mailing list
>>>>> cisco-bba [at] puck
>>>>> https://puck.nether.net/mailman/listinfo/cisco-bba
>>>>
>>>>
>>>>
>>>> --
>>>> No virus found in this incoming message.
>>>> Checked by AVG Anti-Virus.
>>>> Version: 7.0.300 / Virus Database: 265.6.9 - Release Date: 1/6/2005
>>>
>>> ---
>>> Clayton Zekelman
>>> Managed Network Systems Inc. (MNSi)
>>> 344-300 Tecumseh Rd. E.
>>> Windsor, Ontario
>>> N8X 5E8
>>>
>>> tel. 519-985-8410
>>> fax. 519-258-3009
>>>
>>>
>>> --
>>> No virus found in this outgoing message.
>>> Checked by AVG Anti-Virus.
>>> Version: 7.0.300 / Virus Database: 265.6.9 - Release Date: 1/6/2005
>>>
>>
>>
>>
>> --
>> No virus found in this incoming message.
>> Checked by AVG Anti-Virus.
>> Version: 7.0.300 / Virus Database: 265.6.9 - Release Date: 1/6/2005
>
> ---
> Clayton Zekelman
> Managed Network Systems Inc. (MNSi)
> 344-300 Tecumseh Rd. E.
> Windsor, Ontario
> N8X 5E8
>
> tel. 519-985-8410
> fax. 519-258-3009
>
>
> --
> No virus found in this outgoing message.
> Checked by AVG Anti-Virus.
> Version: 7.0.300 / Virus Database: 265.6.9 - Release Date: 1/6/2005
>
>
|
