[Usage]: It seems that vllm doesn't perform well under high concurrency

[syGOAT]

Your current environment

PyTorch version: 2.3.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.5 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: version 3.29.2
Libc version: glibc-2.31

Python version: 3.10.14 (main, Mar 21 2024, 16:24:04) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-91-generic-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA L20
GPU 1: NVIDIA L20
GPU 2: NVIDIA L20
GPU 3: NVIDIA L20
GPU 4: NVIDIA L20
GPU 5: NVIDIA L20
GPU 6: NVIDIA L20
GPU 7: NVIDIA L20

Nvidia driver version: 550.54.14
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.6.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.6.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Byte Order:                         Little Endian
Address sizes:                      52 bits physical, 57 bits virtual
CPU(s):                             180
On-line CPU(s) list:                0-179
Thread(s) per core:                 2
Core(s) per socket:                 45
Socket(s):                          2
NUMA node(s):                       2
Vendor ID:                          GenuineIntel
CPU family:                         6
Model:                              143
Model name:                         Intel(R) Xeon(R) Platinum 8457C
Stepping:                           8
CPU MHz:                            2600.000
BogoMIPS:                           5200.00
Hypervisor vendor:                  KVM
Virtualization type:                full
L1d cache:                          4.2 MiB
L1i cache:                          2.8 MiB
L2 cache:                           180 MiB
L3 cache:                           195 MiB
NUMA node0 CPU(s):                  0-89
NUMA node1 CPU(s):                  90-179
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Unknown: No mitigations
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Mitigation; TSX disabled
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] torch==2.3.0
[pip3] triton==2.3.0
[pip3] vllm-nccl-cu12==2.18.1.0.3.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] nvidia-nccl-cu12          2.20.5                   pypi_0    pypi
[conda] torch                     2.3.0                    pypi_0    pypi
[conda] triton                    2.3.0                    pypi_0    pypi
[conda] vllm-nccl-cu12            2.18.1.0.3.0             pypi_0    pypiROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.4.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     0-89    0               N/A
GPU1    SYS      X      SYS     SYS     SYS     SYS     SYS     SYS     SYS     0-89    0               N/A
GPU2    SYS     SYS      X      SYS     SYS     SYS     SYS     SYS     SYS     0-89    0               N/A
GPU3    SYS     SYS     SYS      X      SYS     SYS     SYS     SYS     SYS     0-89    0               N/A
GPU4    SYS     SYS     SYS     SYS      X      SYS     SYS     SYS     SYS     90-179  1               N/A
GPU5    SYS     SYS     SYS     SYS     SYS      X      SYS     SYS     SYS     90-179  1               N/A
GPU6    SYS     SYS     SYS     SYS     SYS     SYS      X      SYS     SYS     90-179  1               N/A
GPU7    SYS     SYS     SYS     SYS     SYS     SYS     SYS      X      SYS     90-179  1               N/A
NIC0    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS      X 

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

NIC Legend:

  NIC0: mlx5_0

How would you like to use vllm

I started the model:

python -m vllm.entrypoints.openai.api_server --model /root/autodl-tmp/model/Meta-Llama-3-70B-Instruct --tensor-parallel-size 8 --port 8000 --served-model-name gpt-4 --disable-log-stats --trust-remote-code 

My request body is:

{
    "model": "gpt-4",
    "messages": [
        {
            "role": "user",
            "content": "I need feedback on an answer. Please supply the evaluation strictly in JSON format, response format comprising only \"reasoning\" (请用中文简短说明评价的理由，即说明优点也说明缺点，除非实在没有优点或缺点), \"approach\" (请用中文简短说明这个问题应该的回答思路) and \"score\", which should be a number between 1 and 5, indicating the quality of the response. Please note, this is just a customer service response, so there is no need to be overly harsh. However, if the response is less than 5 Chinese characters, the score should be below 3.\nQuestion:MoE 和普通的 LLM 在架构上有什么区别\nAnswer:MoE 有路由和专家，比普通的 LLM 复杂"
        }
    ],
    "temperature": 0.3,
    "top_p": 0.95,
    "max_tokens": 120,
    "guided_json": {
        "type": "object",
        "properties": {
            "reasoning": {
                "type": "string"
            },
            "score": {
                "type": "integer",
                "minimum": 1,
                "maximum": 5
            },
            "approach": {
                "type": "string"
            }
        },
        "required": [
            "reasoning",
            "score",
            "approach"
        ]
    }
}

I conducted a stress test with 20 concurrent users, a runtime of 4 minutes, and a ramp-up time of 1 minute. The performance of vllm was as follows. Total requests 111, API requests per second 0.45, minimum response time 5321ms, average response time 34024ms, maximum response time 53862ms, 90% response time 46041ms, failure rate 3.6%. It seems that vllm doesn't perform well under high concurrency?

[simon-mo]

The performance of guided_json is still a work in progress. Try setting guided_decoding_backend to lm-format-enforcer see whether there's a difference?

[syGOAT]

@simon-mo Unfortunately, its performance was no better after setting guided_decoding_backend to lm-format-enforcer.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!