新川サーバー | エンタープライズ サーバー ハードウェア サプライヤー
Your Storage Cluster Is Bottlenecked at 25GbE. The Next Step Is Not Two More 25GbE Ports. It Is One 100GbE Port.
Adding a second 25GbE port and configuring LACP gives you 50 Gbps aggregate — but no single flow exceeds 25 Gbps. A vMotion migration of a 500 GB VM still takes 3 分. A Ceph recovery after a drive failure still runs at 25 Gbps per node. The Intel E810-CQDA2 is a dual-port 100GbE QSFP28 adapter that delivers 100 ポートあたりの Gbps, line rate, with RDMA support and the same Intel E810 driver stack that VMware, RHEL, and Windows Server have shipped inbox since 2021. It is the NIC that turns your network bottleneck into headroom.
技術仕様
| パラメータ | 仕様 |
|---|---|
| チップセット | Intel Ethernet Controller E810 (Columbiaville) |
| ポート | 2x QSFP28 (100GbE per port), also supports 50GbE, 40運ぶ, 25運ぶ, and 10GbE via breakout/QSFP+ cables |
| ホストインターフェース | PCIe 4.0 x16 (256 Gbps — sufficient for 2x 100GbE bidirectional) |
| RDMA | iWARP and RoCE v2 (configurable in firmware) |
| SR-IOV | まで 256 ポートごとの仮想機能 |
| ADQ (Application Device Queues) | Dedicated hardware queues per application — isolates latency-sensitive workloads at 100GbE |
| DDP (Dynamic Device Personalization) | Reprogram packet processing pipeline for specific protocols |
| ステートレス オフロード | VXLAN, NVGRE, ジュネーブ, IP-in-IP, チェックサム, TSO, LRO at 100GbE line rate |
| 力 | ~14W typical (両方のポートがアクティブです, 100運ぶ, DAC cables) |
| フォームファクタ | Full-height, 半長さ (FHHL) PCIe (x16 physical connector) |
| 物理的な | 167 バツ 111 んん (FHHL PCIe card) |
| ケーブル | QSFP28: DAC passive copper 1-3m, active optical (AOC) 3-30メートル, QSFP28 SR4 optics up to 100m, QSFP28 LR4 optics up to 10km |
| OSサポート | ヴイエムウェア ESXi 7.0 U3+, 8.0+ | Windowsサーバー 2019/2022/2025 | RHEL 8.6+/9.x | SLES 15 SP4+ | Ubuntu 20.04/22.04/24.04 |
| Driver | In-box for VMware ESXi 8.0+, RHEL 8.6+, Ubuntu 22.04+ |
| 互換性のあるサーバー | すべてのデル, xフュージョン, PCIe を搭載した Lenovo サーバー 4.0 x16スロット |
| 保証 | 3-年 (サーバー付き) |
100GbE vs 4x 25GbE LACP — The Single-Flow Performance Gap
| Scenario | 1x 100GbE (E810-CQDA2) | 4x 25GbE LACP | Winner |
|---|---|---|---|
| Single 500 GB VM vMotion | ~40 seconds (single 100 Gbps flow) | ~160 seconds (single flow limited to 25 Gbps) | 100運ぶ (4x faster on single-flow workloads) |
| 10 concurrent 50 GB vMotions | ~40 seconds (10 flows, each gets ~10 Gbps) | ~40 seconds (10 flows distributed across 4x 25GbE ports) | Tie (LACP is effective for multi-flow) |
| Ceph OSD recovery (single PG backfill) | ~40 minutes (100 Gbps single flow) | ~160 minutes | 100運ぶ (single-flow dominates in Ceph) |
| NVMe-oF target (multi-initiator, 8 hosts) | ~95 Gbps aggregate (8 flows) | ~90 Gbps aggregate (8 flows across 4 ポート) | Tie (multi-initiator distributes well) |
The single-flow advantage of 100GbE matters for three specific workloads: vMotion of large VMs, Ceph recovery, and backup/restore of large datasets. For multi-flow workloads, LACP with 4x 25GbE delivers comparable aggregate throughput at roughly the same cost.
互換性のあるプラットフォーム
PCIe 4.0 x16 slot required for full 200 Gbps bidirectional bandwidth. Compatible with Dell R760, R770, R7715, R860, R960, xFusion 2288H V7, 2488H V7, 5885H V7. QSFP28 ports accept QSFP28 DAC, QSFP28 AOC, or QSFP28 SR4/LR4 optics. 100GbE switches: Dell S5232F-ON (32x 100GbE), Arista 7060CX5, Cisco Nexus 9336C-FX2.
新川経由のソース
We supply E810-CQDA2 adapters pre-installed in Dell and xFusion servers. 工場直送価格, 3-年保証, そして世界的な発送.
