Re: PPPoE server (high traffic in WDM network)

Nikos Vassiliadis <[email protected]>
Newsgroups gmane.os.freebsd.isp,gmane.os.freebsd.questions
Message-ID <[email protected]>
> Hello Nikos,

Hi, I just saw your answer while browsing. I am not on isp@...
Please CC questions@.

> Am 2009-07-16 12:27:06, schrieb Nikos Vassiliadis:
>> Michelle Konzack wrote:
>>> I am ongoing to install a CWDM (1GE) and DWDM  (10GE)  network  for  the
>>> Alvarion BreezeACCESS VL (38 base stations) and more then  200  Iskratel
>>> FTTH DSLAMS of 96 ports (each with 100MBit, but only one  1GE  Upstream)
>>> each.
>>
>> So, you'll have 96*200 possible PPP clients. How many concurrent PPP
>> sessions do you care to support?
>> And more importantly, how much aggregate bandwidth?
> 
> Because the customers are permanently On-Line du to the  VoIP-Telephone,
> we count with the full number of clients...
> 
> The distance between the FTTH DSLAM and the customers can be up to 10km.
> 
> The idea is now, that we do not simply connect the FTTH DSLAM's  to  the
> CISCO switches but building a redunant Ethernet Carrier Network.
> 
> This mean, we can install in each village there own FTTH DSLAM  even  if
> there are 2500 hausholds and we install 26 FTTH DSLAM's there.
> 
> This mean in theorie 250 GBit Customer Downstream, 26 Gbit Upstream  but
> we count with a 10 GE which is maybe used to 30-50%.
> 
> OK, if we switch to an "Ethernet Carrier Network" I could install one or
> two PPPoE Servers in each village.  But if one goes down, the second has
> to handel 2500 client connections.

I *think* the number of clients is doable. I don't know about
the bandwidth.

> Note:   This is ONLY the base installation  between  Kehl,  Rheinau,
>         Renchen and Oberkirch (arround  35.000  hausholds)  and  the
>         whole region has 150.000 hausholds.
> 
>> Don't understand what you mean round-robin and loadbalancing?
>> Read below.
> <snip>
>> FreeBSD has a RADIUS library in base. The two notable users of libradius
>> are ppp and net/mpd. The only choice in a ISP environment I think is the
>> net/mpd5 port. Read the outline here:
>> http://www.freebsd.org/cgi/url.cgi?ports/net/mpd5/pkg-descr
>>
>> It is very good and is actually used in large setups.
> 
> Thankyo for the link, I will red on if I am in Office...
> 
>> Can't reply, but keep in mind that filling a 10GE pipe is
>> a hard task on its own.
> 
> It depends on how many customers you have and with an Internet access of
> 100 Mbit plus services like IPTV and VOD you can fill up a 10 GE pipe.

I meant "filling a 10 Gbit pipe with a general purpose computer
architecture is a hard task". Packet forwarding at these rates is
tricky.

>> I *think* having more low fidelity BRASs, will serve your
>> needs better that a few high fidelity ones.
> 
> You mean, putting a bunch of small 1U Servers into a 19" 42RU?

Yes, you may find that having two small boxes instead of bigger one
gives better results performance-wise. You also have to test if SMP
helps and how much. A beast with 16 cores is more powerful from a
regular computer with 2 cores, but does it help in your setup?

>> You can try NanoBSD and TinyBSD which are FreeBSD based and I
>> believe can fit the bill. These two run with their filesystems
>> read-only mounted which is ideal for flash memories.
> 
> Can you recomment it for an ISP setup?

It's FreeBSD running from a read-only mounted medium.
No more, no less. Yes, it's fine for an ISP setup.

> 
> Hmmm, I am right, that NanoBSD can be bootup over network?
> (this would be another solution)

NanoBSD is meant to run in embedded stand-alone devices.
So, I *guess* that is conceptually very far from net booting.

Nikos
_______________________________________________
[email protected] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-isp
To unsubscribe, send any mail to "[email protected]"
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.