{"id":1819,"date":"2024-08-26T10:37:37","date_gmt":"2024-08-26T09:37:37","guid":{"rendered":"https:\/\/blog.thomarite.uk\/?p=1819"},"modified":"2024-11-09T18:02:40","modified_gmt":"2024-11-09T18:02:40","slug":"cloudflare-backbone-2024-cisco-ai-leetcode-alibaba-hpn-altman-ubi-xai-100k-gpu-crowdstrike-rca","status":"publish","type":"post","link":"https:\/\/blog.thomarite.uk\/index.php\/2024\/08\/26\/cloudflare-backbone-2024-cisco-ai-leetcode-alibaba-hpn-altman-ubi-xai-100k-gpu-crowdstrike-rca\/","title":{"rendered":"Cloudflare backbone 2024, Cisco AI, Leetcode, Alibaba HPN, Altman UBI, xAI 100k GPU, Crowdstrike RCA, Github deleted data, DGX SuperPod, how ssh works, Grace Hooper Nvidia"},"content":{"rendered":"\n

Cloudflare backbone 2024<\/a>: Everything very high level. 500% backbone capacity increase since 2021. Use of MPLS + SR-TE. Would be interesting to see how the operate\/automate those many PoPs.<\/p>\n\n\n\n

Cisco AI<\/a>: “three of the top four hyperscalers deploying our Ethernet AI fabric” I assume it is Google, Microsoft and Meta? AWS is the forth and biggest.<\/p>\n\n\n\n

Huawei Cloud Monitor<\/a>: Haven’t read the paper RD-Probe<\/a>. I would expect a git repo with the code \ud83d\ude42 And refers to AWS pdf<\/a> and video<\/a>.<\/p>\n\n\n\n

Automated Leetcode<\/a>: One day, I should have time to use it a learn more programming, although AI can solve them quicker than me \ud83d\ude42<\/p>\n\n\n\n

Alibaba Cloud HPN<\/a>: linkedin<\/a>, paper<\/a>, AIDC material<\/a><\/p>\n\n\n\n

LLM Traffic Pattern: periodically burst flows, few flows (LB harder)<\/p>\n\n\n\n

Sensitive to failures: GPU, link, switch, etc<\/p>\n\n\n\n

Limitations of Traditional Clos: ECMP (hash polarization) and SPOF in TORs<\/p>\n\n\n\n

HPN goals: <\/p>\n\n\n\n

-Scalability: up to 100k GPU<\/p>\n\n\n\n

-Performance: low latency (minimum amount of hops) and maximum network utilization<\/p>\n\n\n\n

-Reliability: Use two TORs with LACP from the host.<\/p>\n\n\n\n

Tier1<\/p>\n\n\n\n

– Use single-chip switch 51.2Tbps. They are more reliable. Dual TOR<\/p>\n\n\n\n

– 1k GPUs in a segment (like nv-link) Rail-optimized network<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

Tier2: Eliminating load imbalance: Using dual plane. It has oversubscription<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n

Tier3: connects several pod. Can reach 100k GPUs. Independent front-end network<\/p>\n\n\n\n

Altman Universal Base Income Study<\/a>: It doesnt fixt all problems, but in my opinion, it helps, and it is a good direction.<\/p>\n\n\n\n

xAI 100k GPU cluster<\/a>: 100k liquid-cooled H100s on single RDMA fabric. Looks like Supermicro involved for servers and Juniper only front-end network. NVIDIA provides all ethernet switches with Spectrum-4. Very interesting. Confirmation<\/a> from NVIDIA (Spectrum used = Ethernet). More details<\/a> with a video.<\/p>\n\n\n\n

Crowdstrike RCA<\/a>:<\/p>\n\n\n\n

Github access deleted data<\/a>: Didn’t know about it. Interesting and scary.<\/p>\n\n\n\n

Nvidia DGX SuperPod: reference architecture<\/a>. video. 1 pod is 16 racks with 4 DGX each (128×8=1024 GPU per pod), 2xIB fabric: compute + storage, fat tree, rail-optimized, liquid cooling. 32k GPU fills a DC.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

How SSH works<\/a>: So powerful, and I am still so clueless about it<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Chips and Cheese GH200<\/a>: Nice analysis for Nvidia Grace CPU (ARM Neoverse) and Hopper H100 GPU<\/p>\n","protected":false},"excerpt":{"rendered":"

Cloudflare backbone 2024: Everything very high level. 500% backbone capacity increase since 2021. Use of MPLS + SR-TE. Would be interesting to see how the operate\/automate those many PoPs. Cisco AI: “three of the top four hyperscalers deploying our Ethernet AI fabric” I assume it is Google, Microsoft and Meta? AWS is the forth and … Continue reading “Cloudflare backbone 2024, Cisco AI, Leetcode, Alibaba HPN, Altman UBI, xAI 100k GPU, Crowdstrike RCA, Github deleted data, DGX SuperPod, how ssh works, Grace Hooper Nvidia”<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13,20,2,1],"tags":[],"class_list":["post-1819","post","type-post","status-publish","format-standard","hentry","category-aws","category-economy","category-networks","category-uncategorised"],"_links":{"self":[{"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/posts\/1819","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/comments?post=1819"}],"version-history":[{"count":6,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/posts\/1819\/revisions"}],"predecessor-version":[{"id":1875,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/posts\/1819\/revisions\/1875"}],"wp:attachment":[{"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/media?parent=1819"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/categories?post=1819"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.thomarite.uk\/index.php\/wp-json\/wp\/v2\/tags?post=1819"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}