Tencent Cloud Account Purchase: Exclusive Advantages of Blackstone Physical Server (BMS)
In the second half of enterprise digitization, "on the cloud" is no longer a new word. However, when the scale of the business moves from rapid trial and error in the early days
Ultra-large-scale data processing, AI large model training, high-concurrency real-time rendering, and core financial transactions.
When it comes to high-performance computing (HPC) scenarios, many technical leaders and architects have fallen into new troubles.
"There are obviously a lot of vCPUs allocated to us. Why does the peak computing power always fail to meet expectations?"
"As soon as the tenants next door run high load, the P99 delay of our service will rise sharply......"
"High-frequency database I/O encounters a bottleneck, and the latency of the cloud disk cannot break through the bottleneck."
The root of these pain points lies in the natural existence of traditional virtualization architecture (Hypervisor)
Power loss, resource contention, and network/storage overhead
.
When the business's demand for performance is approaching the hardware limit, Bare Metal Server (Bare Metal Server) has become the best choice for high-performance computing selection. And in domestic bare metal services,
Tencent Cloud Blackstone Physical Server (BMS)
With its unique architecture design and deep integration capabilities, it shows a very bright "exclusive" advantage.
1. why is high-performance computing so dependent on "physical exclusivity"?
Before talking about Blackstone BMS, we need to clarify:
What does high-performance computing hate about virtualization?
The traditional cloud server (CVM) uses a virtual machine architecture, although it brings a high degree of flexibility and flexibility, but it is separated by a layer of virtualization Hypervisor layer. This mechanism in the face of extreme performance requirements, will expose three fatal short board:
CPU and Memory Loss: Virtualized scheduling consumes additional system resources. In intensive floating-point operations or parallel calculations, frequent context switching of CPU registers and caches (L1/L2/L3 Cache) will lead to invisible loss of up to 5% ~ 15% of computing power.
Noisy Neighbors: Although cloud vendors have an isolation mechanism, if other tenants on the same physical machine suddenly initiate high load on the physical machine in hyperthreading (Hyper-Threading) or bus contention, your CPU cache and memory bandwidth will inevitably be squeezed, resulting in jitter of response delay.
Nested virtualization limitations: If you need to deploy a self-developed container platform, security sandbox, or run a specific deep learning framework in the cloud, secondary virtualization in a virtual machine environment is often extremely low in performance or even impossible to support at all.
Hardcore requirements for high-performance computing: zero virtualization overhead, 100 percent physical resource exclusivity, extreme hardware-level security isolation, and low-to-microsecond network and storage responsiveness.
2. Dismantling Tengxun Cloud Blackstone BMS: Performance of Physical Machines, Experience of Cloud Machines
Tencent Cloud Blackstone Physical Server (Bare Metal Server
, BMS) is the essence
Bare metal cloud products that deeply integrate the "exclusive computing power" of high-performance physical servers with the "elasticity and flexibility" of public cloud computing.
.
It not only retains the loss-free and high-performance hard core attributes of traditional physical machines, but also realizes the ability of minute-level opening and flexible expansion like virtual machines through the hardware offloading architecture (such as smart network card/self-research chip) developed by Tencent Cloud.
Traditional Virtual Machine Architecture Tencent Cloud Blackstone BMS (Bare Metal)
---------------------------------------------------------------------------------------------------------------
│ User Business Application │ │ User Business Application │
├────────────────────────────── ───── ───
│ Virtualization Layer (Hyper) │ <-Loss │ Physical CPU/Memory │ <-100% Lossless Exclusive
├────────────────────────────── ───── ───
│ Physical hardware resources │ │ │ hardware uninstallation (smart network card) │ <-cloud control uninstallation
The --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
From the perspective of technology selection, the core advantages of Blackstone BMS in high-performance computing scenarios can be attributed to the following four dimensions:
1. Extreme computing: zero-loss release of physical CPU and memory potential
Blackstone BMS abandons the Hypervisor layer, and users gain control of the entire physical server.
Physical CPU 100% exclusive: whether it is Intel Xeon high-frequency processor or AMD EPYC multi-core high-performance platform, every core, every thread and all L3 cache of physical CPU are completely exclusive to users' business.
Lossless memory bandwidth: Deep learning and HPC algorithms often rely on extremely high memory throughput rates. The Blackstone physical server is directly connected to the physical NUMA architecture, which avoids the delay loss of virtual machine cross-NUMA access and can push the memory bandwidth to the theoretical limit.
2. Hardware unloading: return the calculation power to the business and hand over the control to the hardware
Traditional bare metal machines often consume physical CPU cores to run network and storage drivers when accessing public networks and cloud storage.
And Tencent Cloud's new generation of Blackstone BMS adopts
Hardware Offload Architecture
. It sinks network virtualization (such as VPC message encapsulation) and storage virtualization to smart network cards (Sma
rtNIC).
Zero CPU usage: 100 percent of the CPU resources of the physical machine are used for the user's computing business and are not occupied by the management process of the cloud platform.
10-million-level pps forwarding: Relying on hardware acceleration, Blackstone BMS can easily achieve up to 10-million-level network packet forwarding performance (pps) and ultra-low network latency, and easily cope with the impact of high concurrent traffic.
3. Seamless integration into cloud ecology: breaking the traditional barrier of "physical machine is isolated island"
The biggest pain point of physical machines hosted by traditional IDC computer rooms is "isolation"-it is difficult to communicate with other services on the public network cloud.
Blackstone BMS completely breaks this barrier:
Native VPC private network isomorphism: Blackstone physical servers naturally run in the VPC (private network) system of Tencent Cloud. This means that your Blackstone physical machine can communicate with common CVM virtual machines, cloud database CDB, and Redis cache clusters in the same VPC in milliseconds.
Cloud hard disk CBS dynamic mount: traditional physical machine expansion storage can only rely on the removal of the machine and add the hard disk. Blackstone BMS has deeply opened up Tengxun Cloud CBS cloud hard disk system, supporting online mounting/unmounting of ultra-high IOPS block storage like virtual machines, and has both physical machine performance and cloud disk elastic scaling and multiple copy disaster tolerance capabilities.
4. Extreme compliance and hardware-level security isolation
For sensitive businesses such as financial transactions, core accounting, and life science genetic data, "sharing physical hardware" is itself a security risk. Blackstone BMS implements
Single-tenant isolation at the physical level
The network, storage and computing are fully independent, meeting the most stringent physical compliance audit requirements.
3. typical landing scene: Blackstone BMS where to show their skills?
The advantages of physical exclusive computing power make Blackstone BMS show irreplaceable value in many industry scenarios where performance is extremely critical:
Industry/Scenario
Core pain point
Blackstone BMS's Breaking Plan
AI Big Models and Deep Learning
The communication delay between training nodes is sensitive, and the GPU/CPU computing power utilization rate is extremely high.
Provides GPU bare metal instances and RDMA high-speed networks to achieve lossless parallel training of ultra-large-scale nodes.
Real-time gaming with high concurrency
Players have zero tolerance for jitter; MMO games are under pressure.
Exclusive CPU main frequency and tens of millions of pps forwarding performance, bid farewell to neighbor interference, and easily guarantee P99 low latency.
Core Database (Oracle/MySQL)
Frequent high concurrent I/O is prone to virtualization disk throughput bottlenecks.
Deployed on high-performance Blackstone nodes, combined with NVMe local disks or extreme CBS, to create a database base with high IOPS and microsecond latency.
Financial Quantitative Trading and Risk Control
The processing delay of trading orders has a demanding microsecond level.
Seek.
Physical hardware is directly connected to the leased line to minimize the queuing time of instructions in the virtualization stack.
4. Conclusion
The selection of high-performance computing is essentially a game of "performance, flexibility and cost.
In the past, enterprises had to choose between the "high elasticity of virtual machines" and the "high performance of traditional physical machines. The emergence of Tencent Cloud Blackstone Physical Server (BMS) has filled the gap between the two through technological innovation:
It enables enterprises to not only have the hard accounting power that is lossless, exclusive and secure for physical machines, but also enjoy the convenience of agile, second-level deployment and rich ecology of public cloud.
If your business is facing the bottleneck of computing power throughput, delay jitter under high concurrency, or strict compliance isolation requirements, you may as well focus on Blackstone BMS-let the core computing power return to physical exclusive, and inject the strongest power into your business engine.
