dokucama

User Tools

Site Tools

Trace:
network_stuff:arista:arista

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
network_stuff:arista:arista [2023/07/29 14:11] – jotasandokunetwork_stuff:arista:arista [2025/10/04 20:42] (current) – [decoding the model code] jotasandoku
Line 1: Line 1:
 +** SWITCHES LINES **
 +
 +  * **7050X3** - 48 x 25 G SFP28 + 8 x 100 G QSFP
 +    * role: leaf / top-of-rack for dense 25 G servers and 100 G uplinks
 +    * asic: Broadcom Trident3
 +
 +  * **7060X5** - 32 x 400 G QSFP-DD (break-outs up to 2 x 800 G)
 +    * role: 400 / 800 G spine for AI-ML clusters and hyperscale clouds
 +    * asic: Broadcom Tomahawk5
 +
 +  * **7060X4** - 32 x 400 G QSFP-DD
 +    * role: 400 G super-spine or fan-out to 4 x 100 G
 +    * asic: Broadcom Tomahawk4
 +
 +  * **7130** - layer-1 matrix, 4 ns port-to-port, FPGA options
 +    * role: ultra-low-latency taps, matrix switches, time stamping
 +    * asic: Xilinx UltraScale+ FPGA cross-point
 +
 +  * **7150** - ~350 ns L2/L3 forwarding
 +    * role: high-frequency trading leaf or edge
 +    * asic: Intel (Fulcrum) Alta FM6000
 +
 +  * **7160-48TC6** - 48 x 10GBASE-T + 6 x 100 G QSFP
 +    * role: high-density 10 G leaf or campus aggregation
 +    * asic: Broadcom Trident3 with AlgoMatch
 +
 +  * **7170** - 32 / 64 x 100 G QSFP with P4-programmable pipeline
 +    * role: telemetry, service chaining, packet engineering
 +    * asic: Intel Barefoot Tofino
 +
 +  * **7280R3** - 32 x 100 G QSFP (deep buffer)
 +    * role: universal leaf / edge router with internet-scale tables
 +    * asic: Broadcom Jericho2
 +
 +  * **7508R3** - modular, up to 288 x 400 G OSFP or QSFP-DD
 +    * role: universal spine or data-centre core up to 230 Tbps
 +    * asic: Broadcom Jericho2C+
 +
 +===== decoding the model code =====
 +
 +  * **number block** – family  
 +    * 70xx fixed switching  
 +    * 72xx deep-buffer fixed  
 +    * 75xx modular chassis  
 +
 +  * **first letter**  
 +    * X - high-performance fixed (Trident / Tomahawk)  
 +    * R - deep-buffer routing (Jericho)  
 +    * S - scaled tables  
 +
 +  * **port letters** (after the number)  
 +    * Q - 40 G QSFP+  
 +    * C - 100 G QSFP28 (or 400 G QSFP-DD when paired with D)  
 +    * D - QSFP-DD 400 G  
 +    * O - OSFP 400 G  
 +    * Y - 25 G SFP28  
 +    * T - 10GBASE-T copper  
 +
 +  * **suffixes**  
 +    * A / F - airflow (A front-to-back, F back-to-front)  
 +    * M - large memory for huge routing tables  
 +    * K - inline MACsec / crypto  
 +    * R3 - third hardware generation  
 +
 +
 +----
 +
 +
 __**ARISTA**__\\ __**ARISTA**__\\
 To enable scp on the management vrf:\\ To enable scp on the management vrf:\\
Line 93: Line 161:
  
 ---- ----
- 
-SWITCHES LINE:\\ 
-  * 7280R : 32 x 100G QSFP  
-    * Universal leaf leaf. Fixed configuration. Up to 100G. ld5 
-  * 7508R3: modular 
-    * For cloud design and datacenter 
-    * 400G OSFP or QSFP-DD  
-  * 7130 : L1 switches with optional FPGAs. As low latency as 4ns. 
-  * 7150 : ultra low latency (400ns) 
-  * 7160-48TC6 
-    * 48 x 10GBASE-T, 6 x 100G QSFP 
- 
  
 Features: Features:
Line 115: Line 171:
   * Media conversion   * Media conversion
   * Telemetry and more   * Telemetry and more
-  * Cut-through technology: We have plain cut-through (6bytes(dmac) enough) cut-through fragment free (64B, not in use anymore). Cut-thru switches can’t check CRC, they will just cause retransmission of bad packets. If the switch uses single-stage crossbar fabric with Virtual Output Queue (VOQ) cut-thru  might not be viable. +  * **Cut-Through Switching (below expanded)** 
 + 
 +Cut-through switching allows forwarding frames as soon as the destination MAC is read (6 bytes), without waiting for the full packet. This enables extremely low-latency forwarding, ideal for HFT environments where every nanosecond counts. 
 +\\ 
 +Switching Modes 
 +  * **Cut-through**: Starts forwarding after 6 bytes (DMAC). Fastest, no CRC check. 
 +  * **Fragment-free**: Waits for 64 bytes to reduce early collisions. Legacy, rarely used. 
 +  * **Store-and-forward**: Waits for full frame and verifies CRC. Safer, slower. 
 +\\ 
 +Trade-offs 
 +  * No CRC/FCS verification – errors propagate to the end host. 
 +  * Risk of retransmission if frames are corrupted. 
 +  * Not always viable with VOQ & crossbar fabrics due to internal queuing or HOL blocking. Virtual Output Queuing prevents head-of-line blocking by organizing separate queues for each destination at every input port. Crossbar fabrics create a matrix architecture where multiple inputs connect simultaneously to multiple outputs. Together, they enable high-throughput, low-latency switching critical for modern data center networks. 
 +\\ 
 +In HFT Environments 
 +  * Critical for market data and order latency minimization. 
 +  * Typically used in clean, controlled networks with reliable links. 
 +  * Acceptable trade-off: CRC skipping vs. latency gains. 
 +\\ 
 +Hardware Notes 
 +  * Arista 7128X, Cisco Nexus 3500, Exablaze support true cut-through. 
 +  * Some switches offer hybrid modes or revert to store-and-forward under congestion. 
 +  * Internal architecture (e.g. VOQ + single-stage fabrics) can limit practical use. 
 +\\ 
 +Summary: Cut-through switching reduces latency by forwarding packets before they are fully received. It skips CRC checks but is widely used in HFT due to its performance gains in controlled, lossless environments.
  
  
Line 170: Line 250:
   * [[https://saidvandeklundert.net/2019-08-06-arista-eapi-using-pyeapi/]]   * [[https://saidvandeklundert.net/2019-08-06-arista-eapi-using-pyeapi/]]
   * [[https://pyeapi.readthedocs.io/en/latest/quickstart.html#connect-to-a-node]]   * [[https://pyeapi.readthedocs.io/en/latest/quickstart.html#connect-to-a-node]]
-  *  
  
 +Using https gives SSLv3 erros with python3.10+. This is to enable higher ciphers and add the certificates: [[https://www.arista.com/en/um-eos/eos-control-plane-security#xx1017549|Link]]
 +\\
 +The quick and dirty solution is to use port :80
 +  ! In the arista side:
 +  management api http-commands
 +    protocol http
 +  username restapi privilege 15 secret restapi
 +  !
 +  In the 'python' side:
 +    import pyeapi
 +  import pprint
 +  eapi_param = pyeapi.client.connect(
 +      transport='http',
 +      host='192.168.121.101',
 +      username='restapi',
 +      password='restapi',
 +      port=80,
 +  )
 +  eapi = pyeapi.client.Node(eapi_param)
 +  version_info = eapi.run_commands(['show version',])
 +  pprint.pprint(version_info)
 +
 +
 +----
 +
 +=== LAYER 1 SWITCHING ===
 +[[https://chatgpt.com/s/dr_68270c44fb0c81919fd19c5ff7fbc074|External Link]]
 +
 +
 +----
 +
 +=== configuration replace ===
 +
 +
 +  scp new.cfg admin@SW1:/mnt/flash/new.cfg    ! scp the new config to the switch
 +  ! ssh to the switch
 +  dir flash:    ! check new.cfg is there
 +  copy running-config flash:prechange.cfg    ! make a checkpoint (golden backup of current running)
 +  wr
 +  reload in 10    ! failsafe
 +  configure replace flash:new.cfg    ! do the config replace
 +  !
 +  ! wait until 9 mins to be sure not locked out due to routing reconverging down thee line!
 +  reload cancel    ! if all looks good, cancel the reload
 +  ! configure replace flash:prechange.cfg     !  only if you need to roll back, restore the checkpoint
network_stuff/arista/arista.1690639867.txt.gz · Last modified: (external edit)

Page Tools

Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 4.0 International
CC Attribution-Share Alike 4.0 International Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki