[AusNOG] Looking for someone who has experience with Nextgen VPLS

Wed Apr 6 19:02:12 EST 2011

Mark, you are correct.

This did not resolve the issue, however it definatelly does not appear to be
an issue with the Sydney<->Perth nextgen link as I can do 100Mbps across it
from a linux box to pretty much anywhere else on my network - however
Windows systems struggle to get even 1Mbps to some locations. I struggle to
understand this behaviour.

We have a win2k3 box in Sydney that only gets 60kb/s cifs speeds to Perth,
while on our previous 8Mbps SHDSL link it was able to get 800KB/s

I've tried putting some old 2811s in between the 2 x 3560's and setting the
adjust-mtu, and window-size options on them, but to no avail.

I'm also being told that due to the latency I need to invest in
TCP/bandwidth accelerators at each end of the link to work around these
issues... I may just be being ignorant about this, but I cant believe this
is the case, and there must be a simple way around this TCP issue that only
appears to affect windows PC's.

Cheers,

Greg

-----Original Message-----
From: Mark Smith
[mailto:nanog at 85d5b20a518b8f6864949bd940457dc124746ddc.nosense.org] 
Sent: Tuesday, 5 April 2011 7:34 PM
To: Greg M
Cc: ausnog at lists.ausnog.net
Subject: Re: [AusNOG] Looking for someone who has experience with Nextgen
VPLS

On Tue, 5 Apr 2011 18:48:10 +0800
"Greg M" <gregm at servu.net.au> wrote:

> Thanks to a couple of replies off-list:
> 
> ip tcp adjust-mss 1452
> 
> One line on our Cisco 3560's resolved everything
> 

That's very strange, because I'm pretty sure that doesn't adjust the TCP
window parameters that usually constrain the amount of data in flight i.e.
the bandwidth delay product. Those are the Window Size field in the TCP
header (limited to 16 bits) and the window scale option, used to scale the
Window Size field up so that it can represent values larger than 65535.
Typically this window size is constrained by how much memory the OS
allocates, with various earlier versions of windows not allocating enough to
fill up these large bandwidth delay product links (also known as LFNs or
Large Fat Networks ('elefants'))

http://en.wikipedia.org/wiki/TCP_window_scale_option

If the above has worked, then maybe the 3560 is adjusting more than the MSS
option, as the MSS option is only sent during the TCP 3 way handshake, well
before Window Sizes for in flight data have any influence.

> Greg
> 
> -----Original Message-----
> From: Mark Smith
> [mailto:nanog at 85d5b20a518b8f6864949bd940457dc124746ddc.nosense.org]
> Sent: Tuesday, 5 April 2011 6:43 PM
> To: Greg M
> Cc: ausnog at lists.ausnog.net
> Subject: Re: [AusNOG] Looking for someone who has experience with 
> Nextgen VPLS
> 
> On Tue, 5 Apr 2011 18:27:11 +0800
> "Greg M" <gregm at servu.net.au> wrote:
> 
> > Thanks again to the members of the Ausnog community the speed issue 
> > is now resolved! :-)
> > 
> 
> Want to summarise what happen between 19:39 and 19:57 tonight? Did you 
> get your Windows TCP windows issues sorted somehow?
> 
> 
> >  
> > 
> > Thanks to everyone who replied,
> > 
> >  
> > 
> > Greg
> > 
> >  
> > 
> > From: ausnog-bounces at lists.ausnog.net 
> > [mailto:ausnog-bounces at lists.ausnog.net] On Behalf Of Greg M
> > Sent: Tuesday, 5 April 2011 6:09 PM
> > To: ausnog at lists.ausnog.net
> > Subject: Re: [AusNOG] Looking for someone who has experience with 
> > Nextgen VPLS
> > 
> >  
> > 
> > Hi All,
> > 
> >  
> > 
> > On the TCP Window size of things. Is there a way for the border 
> > routers/switches on each end of the VPLS to be able to handle this 
> > so that manually setting TCP Window and TCP Window scaling on every 
> > single Windows Desktop PC/server isn't required?
> > 
> >  
> > 
> > I just did some further tests, and from a Centos/Linux virtual 
> > machine in Sydney, I can get 100Mbps to the iiNet FTP, and to both Perth
sites.
> > 
> >  
> > 
> > --2011-04-05 20:03:02--  ftp://ftp.iinet.net.au/test500MB.dat
> > 
> >            => `test500MB.dat'
> > 
> > Resolving ftp.iinet.net.au... 203.0.178.32
> > 
> > Connecting to ftp.iinet.net.au|203.0.178.32|:21... connected.
> > 
> > Logging in as anonymous ... Logged in!
> > 
> > ==> SYST ... done.    ==> PWD ... done.
> > 
> > ==> TYPE I ... done.  ==> CWD not needed.
> > 
> > ==> SIZE test500MB.dat ... 500000000
> > 
> > ==> PASV ... done.    ==> RETR test500MB.dat ... done.
> > 
> > Length: 500000000 (477M)
> > 
> >  
> > 
> > 100%[===============================================================
> > ==
> > ======
> >
> ======================================================================
> ======
> > ================================================>] 500,000,000 
> > 10.1M/s
> in
> > 46s
> > 
> >  
> > 
> > 2011-04-05 20:03:48 (10.3 MB/s) - `test500MB.dat' saved [500000000]
> > 
> >  
> > 
> >  
> > 
> > From a Windows server on the same subnet/switch, I get 4.5MB/s.. So 
> > it seems like the link itself is fine, and it is indeed TCP Window 
> > related. Why this behaviour is so radically different between 
> > Windows and Linux is quite suprising.
> > 
> >  
> > 
> > Thanks,
> > 
> >  
> > 
> > Greg
> > 
> >  
> > 
> > From: ausnog-bounces at lists.ausnog.net 
> > [mailto:ausnog-bounces at lists.ausnog.net] On Behalf Of Greg M
> > Sent: Tuesday, 5 April 2011 4:45 PM
> > To: ausnog at lists.ausnog.net
> > Subject: Re: [AusNOG] Looking for someone who has experience with 
> > Nextgen VPLS
> > 
> >  
> > 
> > Thanks for everyone's replies - I was able to do extensive 
> > troubleshooting today thanks to the excellent replies on and 
> > off-list, and it turns out that there may be multiple issues at 
> > play, including too many MAC's on the circuit, et al.
> > 
> >  
> > 
> > It also seems that Telstra has not enabled 802.1q-in-q on their end 
> > so we are seeing a lot of encapsulation errors, and can only push 
> > traffic across
> > VLAN1 from the Sydney side. They are now telling Nextgen that they 
> > cant enable Q-in-q, which could point to Nextgen provisioning the 
> > wrong type of Telstra service for us:
> > 
> >  
> > 
> > The order from LOLO states "Telstra Wholesale Business Grade 
> > Ethernet Service N275XXXXX for Nextgen Networks"
> > 
> >  
> > 
> > Is this a L3 only service?
> > 
> >  
> > 
> > Since this morning, we have only been able to push 10Mbps through 
> > this VPLS circuit as well, so there definitely seems to be some 
> > faults on the Nextgen/Telstra side of the circuit, even though they 
> > claim everything is configured correctly.
> > 
> >  
> > 
> > Cheers,
> > 
> >  
> > 
> > Greg
> > 
> >  
> > 
> > From: ausnog-bounces at lists.ausnog.net 
> > [mailto:ausnog-bounces at lists.ausnog.net] On Behalf Of Greg M
> > Sent: Monday, 4 April 2011 7:07 PM
> > To: ausnog at lists.ausnog.net
> > Subject: [AusNOG] Looking for someone who has experience with 
> > Nextgen VPLS
> > 
> >  
> > 
> > Hi Noggers,
> > 
> >  
> > 
> > I am looking for someone who is able to consult/provide support to a 
> > large organisation which has recently deployed a 100M Nextgen VPLS 
> > link between an office in Perth & Sydney,  and is experiencing the 
> > most bizzare throughput issues.
> > 
> >  
> > 
> > This link has Nextgen fiber in Perth, with Telstra fiber hand off in 
> > Sydney, resold via Nextgen as an end-to-end VPLS circuit.
> > 
> >  
> > 
> > The issue we face is relating to speed.
> > 
> > -          From Switch A In Sydney to Switch B in Perth, we can do a
> non-df
> > ping of 1500 bytes for 24 hours without a single dropped packet  - 
> > so there are no MTU issues. Latency is also always stable, even when 
> > maxing the link at 100Mbps.
> > 
> > -          Each server in Sydney connected to Switch A, can push the
full
> > VPLS line speed of 100Mbps to servers connected to Switch B in Perth 
> > (two endpoints of the VPLS tails)
> > 
> > -          There is a 3rd site in Perth that has a 1Gbps Amcom fiber
link
> > directly between Switch C in this site, and Switch B in the other 
> > Perth site.
> > 
> > -          Some systems on Switch A in Sydney, can push 100Mbps to some
> > servers on Switch C in Perth site #2, yet others can only push 
> > 10Mbp, while some can only push 1-3Mbps. There are no speed issues 
> > between the two Perth sites, however as they can happily pump a full 
> > 1Gbps to each
> other.
> > 
> > -          There is a 100Mbps link to WAIX connecting into Switch B in
> > Perth. Perth site #2 can do full 100mbps to a test FTP download 
> > (lets say iiNet's 500MB test file on their FTP), yet Sydney can only get
45Mbps.
> > 
> >  
> > 
> > Basically only the directly connected Sydney Switch and Perth switch 
> > are able to max out the 100Mbps VPLS circuit. As soon as traffic 
> > goes beyond this switch, speed seems to degrade - however it doesn't 
> > degrade to all locations, as some systems/servers can obtain a full 
> > 100Mbps transfer speed both inbound and outbound via the Sydney to 
> > Perth VPLS. In Perth sites #1 and #2, there are no speed issues at 
> > all, to any site (there are other remote VPLS and other connected 
> > endpoints), it is only this specific site that is having issues.
> > 
> >  
> > 
> > There are no errors on any of the interfaces in any of the Perth 
> > sites, however the Sydney site reports a lot of Outbound Discards on 
> > the switch port that connects to the Telstra/nextgen VPLS link in
Sydney.
> > 
> >  
> > 
> > Happy to receive responses on or off-list, and we are happy to pay 
> > for someone either locally in Sydney or in Perth to help diagnose 
> > this - as a lot of money is being burnt on this circuit that is not 
> > providing consistent speeds. We have raised these issues to both 
> > Nextgen and Telstra and they are pointing the finger at our internal
network.
> > 
> >  
> > 
> > Thanks!
> > 
> > 
> > Greg
> > 
> 