Nancy:Network: Difference between revisions

From Grid5000
Jump to navigation Jump to search
No edit summary
 
(105 intermediate revisions by 13 users not shown)
Line 1: Line 1:
{{Template:Site link|Network}}
{{Template:Site link|Network}}
{{Portal|Network}}
{{Portal|Network}}
{{Status|In production}}
{{Portal|User}}
{{Maintainer|Philippe Robert}}


= IP networks in use =
'''See also:''' [[Nancy:Hardware|Hardware description for Nancy]]


You have to use a public network range to run an experiment between several Grid5000 sites.
= Overview of Ethernet network topology =


=== Public Networks ===
[[File:NancyNetwork.svg|800px]]


* computing : '''172.28.52.0/22'''
{{:Nancy:GeneratedNetwork}}
* virtual : '''10.144.0.0/14'''


=== Local Networks ===
= HPC Networks =


* ipmi : '''172.28.152.0/22'''
Several HPC Networks are available.
* ib : '''192.168.210.0/24''' : see [[Nancy:Network#High_Performance_Networks|High Performance Networks]]


= Network =
== Omni-Path 100G on grele and grimani nodes  ==


== Production Network ==
*<code class="host">grele-1</code> to <code class="host">grele-14</code> have one 100GB Omni-Path card.
*<code class="host">grimani-1</code> to <code class="host">grimani-6</code> have one 100GB Omni-Path card.


[[Image:NetworkNancy.png|Production network|902px]]
* Card Model: Intel Omni-Path Host Fabric adaptateur series 100 1 Port PCIe x8
[[Image:NetworkNancy2.png|Production network]]


== Physical link details ==
== Omni-Path 100G on grvingt nodes  ==


The following diagram show a view of the central router, a hp procurve 5406zl, named sgravillon1.  
There's another, separate Omni-Path network connecting the 64 grvingt nodes and some servers. The topology is a non blocking fat tree (1:1).
* All internal links are 10GB CX4.
Topology, generated from <code>opareports -o topology</code>:
* Renater link is 10GB optical fiber.


[[Image:routeur-nancy2.png|600px]]
[[File:Topology-grvingt.png|400px]]


More information about using Omni-Path with MPI is available from the [[Run_MPI_On_Grid%275000]] tutorial.


<table class="program">
'''NB: OPA (Omni-Path Architecture) is currently not supported on Debian 12 environment.'''
<tr>
    <th>A1</th>
    <th>A3</th>
    <th>A5</th>
    <th>A7</th>
    <th>A9</th>
    <th>A11</th>
    <th>A13</th>
    <th>A15</th>
    <th>A17</th>
    <th>A19</th>
    <th>A21</th>
    <th>A23</th>
</tr>
<tr>
    <td>fgravillon2-NIC1</td>
    <td>Nc (fgravillon1-NIC2)</td>
    <td>fgriffon3-ipmi</td>
    <td>fgriffon4-ipmi</td>
    <td>fgrelon1-NIC2</td>
    <td>fgriffon1-Prod (fgrelon2-NIC1)</td>
    <td>sgrenade1</td>
    <td>sgrenade3</td>
    <td>sgrenade5</td>
    <td>sgriffon2-ipmi</td>
    <td>sgriffon3-ipmi</td>
    <td>fgriffon1-Mgt</td>
</tr>
<tr>
    <th>A2</th>
    <th>A4</th>
    <th>A6</th>
    <th>A8</th>
    <th>A10</th>
    <th>A12</th>
    <th>A14</th>
    <th>A16</th>
    <th>A18</th>
    <th>A20</th>
    <th>A22</th>
    <th>A24</th>
</tr>
    <td>fgriffon2-ipmi</td>
    <td>Nc (fgravillon1-NIC1)</td>
    <td>fgravillon2-NIC2</td>
    <td>fgrelon1-NIC1</td>
    <td>fgriffon1-ipmi</td>
    <td>Nc (fgrelon2-NIC2)</td>
    <td>sgrenade2</td>
    <td>sgrenade4</td>
    <td>LORIA</td>
    <td>sgriffon1-ipmi</td>
    <td>fgriffon3-eth0</td>
    <td>fgriffon3-eth1</td>
</table>


== High Performance Networks ==
=== Switch ===
 
=== Infiniband 20G on griffon nodes  ===
 
==== Nodes ====
 
*<code class="host">griffon-1</code> to <code class="host">griffon-92</code> have one 20GB Infiniband card.
 
* Card Model : Mellanox Technologies MT26418 [ConnectX IB DDR, PCIe 2.0 5GT/s] ( [http://www.mellanox.com/content/pages.php?pg=products_dyn&product_family=4&menu_section=41 ConnectX] ).
* Driver : <code class="dir">mlx4_ib</code>
* OAR property : ib20g=YES ( [https://helpdesk.grid5000.fr/oar/Nancy/monika.cgi?props=ib20g%3DYES&Action=Display+nodes+for+these+properties&.cgifields=props Monika] ).
* IP over IB addressing : <code class="host">griffon-[1..92]-ib</code>.nancy.grid5000.fr ( 192.168.210.[1..92] )
 
==== Switch ====


* Infiniband Switch 4X DDR
* Infiniband Switch 4X DDR
* Model based on [http://www.mellanox.com/content/pages.php?pg=products_dyn&product_family=16&menu_section=33 Infiniscale_III]
* Model based on [http://www.mellanox.com/content/pages.php?pg=products_dyn&product_family=16&menu_section=33 Infiniscale_III]
* 1 commutation card Flextronics F-X43M204
* 1 commutation card Flextronics F-X43M204
* 11 line cards 4X 12 ports DDR Flextronics F-X43M203
* 12 line cards 4X 12 ports DDR Flextronics F-X43M203
 
==== Interconnection ====
 
Infiniband network is physically isolated from Ethernet networks. Therefore, Ethernet network emulated over Infiniband is isolated as well. There isn't any interconnexion, neither at the data link layer nor at the network layer.
 
=== Myrinet 10G on griffon nodes ===
 
==== Nodes ====
 
*<code class="host">griffon-11</code> to <code class="host">griffon-14</code> have one 10GB Myrinet card.
 
* Card Model : Myrinet 10G ( [http://www.myri.com/Myri-10G/10gbe_solutions.html] ) 10G-PCIE-8B-C NIC
* Driver : <code class="dir">myri10ge</code>
* OAR property : ib20g=YES ( [https://helpdesk.grid5000.fr/oar/Nancy/monika.cgi?props=eth10g%3DYES&Action=Display+nodes+for+these+properties&.cgifields=props Monika] ).
* Ethernet. IP addresses : <code class="host">griffon-[11..14]-myri0</code>.nancy.grid5000.fr ( 172.28.55.[221..224] )
 
==== Interconnection ====
 
The nodes having myrinet configured are connected to the 10G Ethernet interfaces of the switches sgriffon|1-3] as following:
* griffon-11 to sgriffon1
* griffon-12 to sgriffon1 (the configuration is not working probably due to the switch configuration)
* griffon-13 to sgriffon2
* griffon-14 to sgriffon3
 
= Grid5000 interconnect =
== Interconnect type ==
[[Image:nancy_g5k_interconnect.png|Nancy Grid'5000 interconnect]]
 
== Link details ==
[[Image:nancy_g5k_linking.png|Nancy's Grid'5000 linking]]
 
'''Note''': All the used fiber cables are dedicated to our Grid'5000 interconnect. When it is said ''trunk'' on the above figure it means a ''more rigid garter''.


= Loria interconnect =
=== Interconnection ===
== Interconnect type ==
[[Image:nancy_loria_interconnect.png|Nancy's Loria interconnect]]


== Link details ==
Omnipath network is physically isolated from Ethernet networks. Therefore, Ethernet network emulated over Infiniband is isolated as well. There isn't any interconnexion, neither at the L2 or L3 layers.
[[Image:nancy_loria_linking.png|Nancy's Loria linking]]

Latest revision as of 07:22, 19 June 2024

See also: Hardware description for Nancy

Overview of Ethernet network topology

NancyNetwork.svg


Network devices models

  • gw-next: Aruba 8325-48Y8C JL635A
  • gw: Cisco Nexus 9508
  • sgrappe: Dell S5224F-ON
  • sgrele-opf: Omni-Path
  • sgros1: Dell Z9264F-ON
  • sgros2: Dell Z9264F-ON
  • sgruss: Dell S5224F-ON
  • sgrvingt: Dell S4048

More details (including address ranges) are available from the Grid5000:Network page.

HPC Networks

Several HPC Networks are available.

Omni-Path 100G on grele and grimani nodes

  • grele-1 to grele-14 have one 100GB Omni-Path card.
  • grimani-1 to grimani-6 have one 100GB Omni-Path card.
  • Card Model: Intel Omni-Path Host Fabric adaptateur series 100 1 Port PCIe x8

Omni-Path 100G on grvingt nodes

There's another, separate Omni-Path network connecting the 64 grvingt nodes and some servers. The topology is a non blocking fat tree (1:1). Topology, generated from opareports -o topology:

Topology-grvingt.png

More information about using Omni-Path with MPI is available from the Run_MPI_On_Grid'5000 tutorial.

NB: OPA (Omni-Path Architecture) is currently not supported on Debian 12 environment.

Switch

  • Infiniband Switch 4X DDR
  • Model based on Infiniscale_III
  • 1 commutation card Flextronics F-X43M204
  • 12 line cards 4X 12 ports DDR Flextronics F-X43M203

Interconnection

Omnipath network is physically isolated from Ethernet networks. Therefore, Ethernet network emulated over Infiniband is isolated as well. There isn't any interconnexion, neither at the L2 or L3 layers.