SURFnet6 tests

As part of our 2005 workplan we intend to perform network tests on the new SURFnet6 infrastructure.
We are cooperating with CWI.

We intend to keep a log of the activities in this Wiki server.

The actual web pages with the results is:
http://vangogh0.uva.netherlight.nl/SURFnet6Tests∞ (still being extended)

Phase 1. Tests with CWI

We have connected one of our Rembrandt nodes (rembrandt2 currently) directly to the CWI Foundry NetIron 40G, at 10 GE.
CWI has one node with 4 1GE interfaces connected to the same switch.
The tests we will perform are:

• Bandwitdh measurements with iperf, using UDP and TCP
• Multicast: ssmping and iperf
• Rate limiting tests

We intend to repeat the above tests with:

• Jumbo frames and standard Ethernet MTUs (9000bytes and 1500bytes, respectivly)
• IPv4 and IPv6 addresses

Results phase1 notes

Apr. 28 2005.

We have started to run a first series of tests between UvA equipment and CWI equipment.
1. Tests between two hosts configured for Jumbo frame do not run. Only when we select a value of 1500bytes for the MTU tests run...
Questions: is the 10GE interface toward the UvA node configured for Jumbo frames?

2.The maximum achieved throughout is ~650mbps. The CWI nodes have 4 1gig link bundled together
Question: is the Foundry still only pushing traffic onto one link?
JP mentioned the Foundry would load balance on multiple links, if one full. But this does not seem the case.
Where the packets are going through should be checked through monitoring the Foundry SNMP counters.

Phase 2. Tests through SURFnet6 Layer 2 network

Jun. 26 2005 (updated: Jun 30th)

SURFnet will hookup the AIR Nortel to their Nortel this week and assign a tagged VLAN id (20050729: 3000) which will be placed on the AIR (*untagged*) and CWI (*tagged*) ports to facilitate the L2 testing.
Apart from that SURFnet will also setup the tagged VLAN id's for AIR to get to the SARA (Arthur) and TC2 (Trillian) Avici's.
CWI has requested an additional IPv4 /24 for the L3 testing (just to have all the ingredients ready), currently there is 194.171.32.0/24 for testing, hopefully we can add 194.171.33.0/24 (20050728: assigned to CWI) to that.

CWI's port facing AIR has been enabled and placed into VLAN 3000 (untagged), the port facing SURFnet6 has also been placed into VLAN 3000 (tagged). Everything should now be done through the glimmerglass (speculaas) probably.

Note: JP is not at CWI much the whole month of august, though is reachable through e-mail and phone (4125) and has made room for the SURFnet6 tests during his semi downtime ;)

Aug.03 2005 by Paola
The setup CWI <-> SURFnet6 ERS <-> Uva ERS is up and functional.
It took some time to have it working. A couple of TX/RX fibers swapped, but the no path between the two test machines (right now rembrandt6 and rembrandt7) until an IP address was configured on the Foundry. Was this the cause of the problems?

Sep.06 2005 by JP
All of the weird asymmetric problems we experienced yesterday (in what seemed like a Foundry melting) have disappeared, tests run during the night showed a steady flow of 6Gbps for several hours.

Sep.07 2005 by JP
A little infrastructure update. Hopefully at the end of the afternoon or otherwise sometime tomorrow, CWI will have a fully functional set of 4 pair of SingleMode and 8 pair of MultiMode fibers going from the CWI M381 room towards SARA comms-room S043. Also, we've been given 2 extra 10GE LANPHY XENPAK's on loan by Foundry, bringing the total number of usable 10GE ports on our production chassis to 4.

Sep.21 2005 by JP
Another update, we've seen issues where having the "flow-control" enabled on the CWI Foundry SURFnet6 port would influence the setup where apparently when the Foundry send it PAUSE frame that triggered a 30 seconds and upwards stall in the path AIR-SURFnet6-CWI. Based on later information (mails with Andree at SARA) and simple testing (int up/down) it would appear is if something very much akin to an STP recalculation occured at the SURFnet6 end.

We have been unable to recreate this incidents so far. Foundry TAC is looking into the issue and has reluctantly given the command 'dm status rmon 1/1' to make the NI40G give some more detailed information about packet counters.

Sep.21 2005 @ 22:50 by JP
The 4x10GE linecard on the CWI Foundry NI40G crashed during a benign 'show' command. Foundry TAC is investigating...

---

AIR / SURFnet6 / CWI testing scenarios

to be done by JP, with Marijke Colors [TM]

http://homepages.cwi.nl/~jpv/SURFnet6/AIR-SURFnet6-CWI-switchrouter.html∞


Nov.3 2005 by JP
Today Paola and JP put in the physical fiber and optics in order to get a 1GbE link from the AIR DAS2 Netgear switch up to the CWI Foundry NI40G, this in preparation to have the DAS2 nodes be able to blast large amounts of low traffic streams to eachother through several paths.

Nov.7 2005 by JP
SURFnet contacted CWI that they formally have asked SARA to create a second 10GE connection. This one from the *new* SURFnet6 testbed to the CWI Foundry NI40G, augmenting the existing one to the production SURFnet6 network (which is in use for production traffic, next to being subjected to AIR+CWI tests).

Nov.24 2005 by JP
Requested the extra IPv4 /24 for the DAS2 flows experiment (assigned 2005/11/28), put things in motion to bug SARA some more about moving the CWI production SURFnet6 link onto the CWI owned fibertrunk (pair CWI/M381/PATCH0-SARA/S043/AA10-SM-01), also got word that SURFnet bugged SARA again about setting up the CWI SURFnet6 testbed link on another pair of that fibertrunk (pair CWI/M381/PATCH0-SARA/S043/AA10-SM-02). Once both actions are done and working OK, we can re-use the fiber going from CWI into NIKHEF to get another 10GE link directly between UvA LightHouse (Glimmerglass) and CWI's Foundry NI40G.

The end result will hopefully be:

• CWI:
  • 2 10GE links towards AIR / LightHouse Rembrandt Nodes
  • 10GE towards SURFnet6 testbed
  • 10GE towards SURFnet6 production
  • GE towards DAS2 Netgear
• AIR:
  • 10GE towards SURFnet6 testbed
  • 10GE towards SURFnet6 production (hopefully)
  • GE towards DAS2 Netgear

Dec.06 2005 by JP
SARA repatched the CWI SURFnet6 production fiber route onto pair CWI/M381/PATCH0-SARA/S043/AA10 SM01-04 pair 03. Supposedly pair 04 will be connected to the SURFnet6 testbed. As for the connections towards the UvA LightHouse, pair 01 has been physically looped at the CWI end, which seems to work when connecting Rembrandt1 to it through the Glimmerglass (port 31), pair 02 however seems to have problems. That pair is connected to the CWI Foundry NI40G on eth1/4 and has been looped on the Glimmerglass (port 32), yet no link is seen (behaviour did occur when looping port 18 which is connected on eth1/3 !). SARA has been contacted in order to investigate if this is a problem at either CWI or UvA, or somewhere in the fiberpath...
As for connecting the AIR Nortel ERS8600R to the SURfnet6 testbed, word from SURFnet is that once CWI has a working connection to the SURFnet6 testbed Nortel ERS8600R and all the VLAN's have been configured, the AIR group will lose its connection to the SURFnet6 production Nortel ERS8600R and gain one on the testbed. This will depend on whether the SURFnet6 NOC can prepare the testbed for the CWI and AIR tests...