[yoursunny]

MTU is good now (except MASS).
I made a slice in every available location with FABNetv4 network service, tested ping with a few MTUs (256, 1280, 1420, 1500, 8900, 8948, 9000).
They can all support MTU 8948 (IPv4 ping -s 8920), but not MTU 9000 (IPv4 ping -s 8972).

IPv4 ping MTU and RTT
src\dst  |   CERN   |   UCSD   |   DALL   |   NCSA   |   CLEM   |   TACC   |   MAX    |   WASH   |   GPN    |   INDI   |   FIU    |   MICH   |   MASS   |   UTAH   |   SALT   |   STAR  
---------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------|----------
CERN     | 9000   0 | 8948 148 | 8948 122 | 8948 105 | 8948 105 | 8948 128 | 8948  91 | 8948  88 | 8948 156 | 8948 107 | 8948 115 | 8948 108 | 1420 101 | 8948 133 | 8948 133 | 8948 102
UCSD     | 8948 148 | 9000   0 | 8948  43 | 8948  48 | 8948  76 | 8948  49 | 8948  62 | 8948  59 | 8948  37 | 8948  50 | 8948  86 | 8948  50 | 1420  72 | 8948  14 | 8948  14 | 8948  45
DALL     | 8948 122 | 8948  43 | 9000   0 | 8948  22 | 8948  50 | 8948   5 | 8948  36 | 8948  34 | 8948  51 | 8948  24 | 8948  60 | 8948  25 | 1420  46 | 8948  28 | 8948  28 | 8948  19
NCSA     | 8948 105 | 8948  48 | 8948  22 | 9000   0 | 8948  32 | 8948  28 | 8948  19 | 8948  16 | 8948  56 | 8948   7 | 8948  43 | 8948   7 | 1420  29 | 8948  33 | 8948  33 | 8948   2
CLEM     | 8948 105 | 8948  76 | 8948  50 | 8948  32 | 9000   0 | 8948  56 | 8948  19 | 8948  16 | 8948  84 | 8948  35 | 8948  43 | 8948  35 | 1420  28 | 8948  61 | 8948  61 | 8948  30
TACC     | 8948 128 | 8948  49 | 8948   5 | 8948  28 | 8948  56 | 9000   0 | 8948  42 | 8948  39 | 8948  57 | 8948  30 | 8948  66 | 8948  30 | 1420  52 | 8948  34 | 8948  34 | 8948  25
MAX      | 8948  91 | 8948  62 | 8948  36 | 8948  19 | 8948  19 | 8948  42 | 9000   0 | 8948   2 | 8948  70 | 8948  21 | 8948  29 | 8948  22 | 1420  15 | 8948  47 | 8948  47 | 8948  17
WASH     | 8948  88 | 8948  59 | 8948  34 | 8948  16 | 8948  16 | 8948  39 | 8948   2 | 9000   0 | 8948  67 | 8948  18 | 8948  26 | 8948  19 | 1420  12 | 8948  44 | 8948  44 | 8948  14
GPN      | 8948 156 | 8948  37 | 8948  51 | 8948  56 | 8948  84 | 8948  57 | 8948  70 | 8948  67 | 9000   0 | 8948  58 | 8948  94 | 8948  58 | 1420  80 | 8948  22 | 8948  23 | 8948  53
INDI     | 8948 107 | 8948  50 | 8948  24 | 8948   7 | 8948  35 | 8948  30 | 8948  21 | 8948  18 | 8948  58 | 9000   0 | 8948  45 | 8948   9 | 1420  31 | 8948  35 | 8948  35 | 8948   4
FIU      | 8948 115 | 8948  86 | 8948  60 | 8948  43 | 8948  43 | 8948  66 | 8948  29 | 8948  26 | 8948  94 | 8948  45 | 9000   0 | 8948  46 | 1420  39 | 8948  71 | 8948  71 | 8948  40
MICH     | 8948 108 | 8948  50 | 8948  25 | 8948   7 | 8948  35 | 8948  30 | 8948  22 | 8948  19 | 8948  58 | 8948   9 | 8948  46 | 9000   0 | 1420  31 | 8948  36 | 8948  35 | 8948   5
MASS     | 1420 101 | 1420  72 | 1420  46 | 1420  29 | 1420  28 | 1420  52 | 1420  15 | 1420  12 | 1420  80 | 1420  31 | 1420  39 | 1420  31 | 9000   0 | 1420  57 | 1420  57 | 1420  26
UTAH     | 8948 133 | 8948  14 | 8948  28 | 8948  33 | 8948  61 | 8948  34 | 8948  47 | 8948  44 | 8948  22 | 8948  35 | 8948  71 | 8948  36 | 1420  57 | 9000   0 | 8948   0 | 8948  30
SALT     | 8948 133 | 8948  14 | 8948  28 | 8948  33 | 8948  61 | 8948  34 | 8948  47 | 8948  44 | 8948  23 | 8948  35 | 8948  71 | 8948  35 | 1420  57 | 8948   0 | 9000   0 | 8948  30
STAR     | 8948 102 | 8948  45 | 8948  19 | 8948   2 | 8948  30 | 8948  25 | 8948  17 | 8948  14 | 8948  53 | 8948   4 | 8948  40 | 8948   5 | 1420  26 | 8948  30 | 8948  30 | 9000   0

[yoursunny]

MTU issue is discovered between MASS and STAR on the experiment network.
I increased MTU of every netif to 9000, but the largest IPv4 ping that can pass through is 1424.
Slice ID: 3b8d1e30-8c17-45b2-9e78-4e59f69cfc3e

ubuntu@NA:~$ ping -M do -c 4 -s 1424 192.168.8.2
PING 192.168.8.2 (192.168.8.2) 1424(1452) bytes of data.
1432 bytes from 192.168.8.2: icmp_seq=1 ttl=64 time=26.8 ms
1432 bytes from 192.168.8.2: icmp_seq=2 ttl=64 time=26.7 ms
1432 bytes from 192.168.8.2: icmp_seq=3 ttl=64 time=26.7 ms
1432 bytes from 192.168.8.2: icmp_seq=4 ttl=64 time=26.7 ms

--- 192.168.8.2 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3005ms
rtt min/avg/max/mdev = 26.659/26.695/26.765/0.041 ms
ubuntu@NA:~$ ping -M do -c 4 -s 1425 192.168.8.2
PING 192.168.8.2 (192.168.8.2) 1425(1453) bytes of data.

--- 192.168.8.2 ping statistics ---
4 packets transmitted, 0 received, 100% packet loss, time 3050ms

 

[yoursunny]

I’m seeing MTU issues on the data plane network between SALT and UTAH.
The scenario is using NIC_ConnectX_5 NICs and L2PTP network service.

I increased the MTU of VLAN netifs to 9000, and assigned IPv4 addresses to both ends.
The maximum ICMP ping size that can pass through is 1472.

ubuntu@9bf529e4-3efc-4988-a9f8-5089ccfa08af-nb:~$ ping -M do -c 4 -s 1472 192.168.8.1
PING 192.168.8.1 (192.168.8.1) 1472(1500) bytes of data.
1480 bytes from 192.168.8.1: icmp_seq=1 ttl=64 time=0.348 ms
1480 bytes from 192.168.8.1: icmp_seq=2 ttl=64 time=0.225 ms
1480 bytes from 192.168.8.1: icmp_seq=3 ttl=64 time=0.227 ms
1480 bytes from 192.168.8.1: icmp_seq=4 ttl=64 time=0.192 ms

--- 192.168.8.1 ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3051ms
rtt min/avg/max/mdev = 0.192/0.248/0.348/0.059 ms

ubuntu@9bf529e4-3efc-4988-a9f8-5089ccfa08af-nb:~$ ping -M do -c 4 -s 1473 192.168.8.1
PING 192.168.8.1 (192.168.8.1) 1473(1501) bytes of data.

--- 192.168.8.1 ping statistics ---
4 packets transmitted, 0 received, 100% packet loss, time 3074ms

[yoursunny]

there is no way to upload files from local to Fabric nodes directly

It’s possible in two ways:

• Host your file with an HTTPS server somewhere on the Internet (with HTTP Basic authentication if desired), and download it on the nodes with wget command.
• Add the nodes into your local ~/.ssh/config with ProxyJump through the bastion, and then run scp to upload the file to the nodes.

I’ve done both in different experiments, but only the first one can be automated.

[yoursunny]

NIC_Basic is a Virtual Function (VF) on the ConnectX-6 Ethernet adapter. The hardware Ethernet adapter is shared among many VFs, and it determines which VF shall receive an incoming packet by matching the destination address. Therefore, NIC_Basic cannot receive Ethernet frames whose destination address differs from its own address.

[yoursunny]

it seems as if packets are filtered by MAC learning on the L2Bridge type network

What observation led you to this conclusion?

What are you trying to do, how it behaved, and how do you expect it to behave?

[yoursunny]

When you invoke slice.submit(), the slice object (and the associated nodes, links, intfs) will not auto-update.
You need to do another slice = fablib.get_slice(name=slice.get_name()) (and re-retrieve enclosed nodes, link, intfs if needed) to obtain updated information.

[yoursunny]

I just noticed this:

Since FABLib v1.4, API documentation is hosted at https://fabric-fablib.readthedocs.io/.
Older version of FABLib API documentation is at https://learn.fabric-testbed.net/docs/fablib/fablib.html.

Can you add a note on top of the “older version” page, pointing to the v1.4 docs?
I have bookmarked the “older version” page previously, and have been wondering why it doesn’t update, until now.
I suppose some others may have done the same.

[yoursunny]

ifconfig is deprecated since Ubuntu 16.

Try ip link or ip addr command.

[yoursunny]

I faced the same issue today.
I found this method effective:

1. File – Hub Control Panel
2. Stop My Server
3. Logout
4. Login again

YMMV

[yoursunny]

Ilya Baldin wrote:

I know for a fact we have folks using DPDK on the testbed with all three NIC types

Yeah, that’s me.
All three NIC types can support most DPDK features, but for NIC_Basic you must use the assigned MAC address and cannot use other MAC addresses.

You would need linux-image-generic for the kernel module and libibverbs-dev to enable DPDK net_mlx5 driver.
In FABlib script, you can insert this step, after slice.submit() and before building DPDK:

execute_threads = {}
for node in slice.get_nodes():
    execute_threads[node] = node.execute_thread(f'''
        sudo apt update
        sudo DEBIAN_FRONTEND=noninteractive apt full-upgrade -y
        sudo DEBIAN_FRONTEND=noninteractive apt install -y –no-install-recommends libibverbs-dev linux-image-generic jq
        sudo reboot
    ''')
for thread in execute_threads.values():
    thread.result()
slice.wait_ssh(progress=True)

[yoursunny]

I see indentation errors in the two snippets above.

Also, there’s no Preview button so that it’s impossible to know whether my content would render correctly before publishing.

[yoursunny]

We have on the roadmap support for ERO (explicit route objects) so that you can have your traffic take the ‘scenic route’ as you say

This will be very useful.

You can on your own add delay to your traffic using traditional Linux tc tools.

This would not work well in speeds higher than 10 Gbps.

running Iperf I saw the max bandwidth I can get in VMs is around 19-20 Gbps

You might get higher speeds if you request dedicated NIC.
I can get 60 Gbps on ConnectX-6 dual-port 100 Gbps and 33 Gbps on ConnectX-5 dual-port 25 Gbps, tested in SALT location.
This number is “goodput” reported by my DPDK-based program, and it’s the sum of traffic in both directions.
iperf3 is unidirectional traffic so it could be half of this.

[yoursunny]

I agree that knowing the physical topology may be useful, especially when the slice spans multiple sites.
For example, if I create a slice with L2PTP link between STAR and UCSD, there are multiple possible paths, and it’s good to know which path my traffic is taking.

Add onto this, it’s useful to have observability into other traffic on the same links.
The link between WASH and STAR has 100 Gbps capacity.
When I benchmark my app running over L2PTP link between these two sites, I may receive throughput of 60 Gbps.
This could be caused by issues in my application (e.g. congestion control algorithm not tuned for >10ms RTT), or caused by other competing traffic on the link.
If FABRIC portal can include a diagram of near-realtime link utilization, I can better understand which one is the more likely cause.

In another angle, it’s also useful to be able to control the physical topology.
Suppose I want to develop my congestion control algorithm over a high latency link, the longest RTT can be achieved today is between MASS and UCSD.
If I can specify the physical topology as part of slice definition, I could make a link to take a scenic route such as WASH-ATLA-DALL-LOSA-SALT-KANS-STAR, which provides a longer natural latency.

[yoursunny]

Can you enable Markdown too?
Both GitHub and Jupyter have Markdown, so more developers are already familiar with it.

[Author]

Posts