Customizing the Server Baseline for Comparison and Diagnosis
You can customize a baseline file as the server performance benchmark. Then you can use the Kunpeng Health Inspector to collect the actual running data of the server. It diagnoses the server and determines its status by comparing it with the baseline.
Prerequisites
The Kunpeng Health Inspector is installed in /home/devkit-kspect-x.x.x-Linux-aarch64.
Procedure
- Go to the tool directory.
cd /home/devkit-kspect-x.x.x-Linux-aarch64
- View the baseline template file.
ll documents
Command output:
total 32 -rw-------. 1 root root 12519 Feb 24 20:04 base.xlsx -rw-------. 1 root root 9005 Feb 24 20:04 collect_commands.txt -rw-------. 1 root root 458 Feb 24 20:04 server_config.ini
- Edit the baseline file.
You can use Excel to edit the base.xlsx file and write the baseline file based on the usage description in base.xlsx. The following is a content example of the baseline file for UnifiedBus:
Fill in the table as required. If the impact or suggested modification is not provided, it will appear as -- in the comparison report where inconsistencies exist.
Path
Baseline Value
Collection Item
Impact
Suggested Modification
bios->BIOS detail Info->LPI
Disabled
KVCache Performance-CPU PM Contrl
LPI is used to optimize power consumption management and resource allocation. Disabling LPI can improve CPU performance.
UB->KVCache Info->Enable Urma
TRUE
KVCache Performance-enable_urma
The datasystem worker components pull data from each other through the URMA API.
UB->KVCache Info->Enable Rdma
TRUE
KVCache Performance-enable_rdma
UB->KVCache Info->Urma Mode
UB
KVCache Performance-urma_mode
The datasystem worker components pull data from each other through URMA and UnifiedBus, improving the performance of cross-node data pulling of KV cache.
UB->KVCache Info->Enable Worker Worker Batch Get
TRUE
KVCache Performance-enable_worker_worker_batch_get
The datasystem worker components batch pull data from each other, improving the performance of cross-node data pulling of KV cache.
memory->OS Memory Info->Automatic NUMA Balancing
Disabled
UBS Comm-NUMA balancing
NUMA balancing must be disabled to avoid interference with Unified-based communication.
UB->UBSVirt Info->Pod Cpu Config
[50,]m
UBSVirt Container Network-CPU Baseline
Setting cpu>=50m is recommended; otherwise, DP container performance and stability may be affected.
UB->UBSVirt Info->Pod Memory Config
[50,]Mi
UBSVirt Container Network-Memory Baseline
Setting memory>=50Mi or leaving it unspecified is recommended; otherwise, DP container performance and stability may be affected.
UB->UBSVirt Info->Dp Pod Status
running
UBSVirt Container Network-DP Container Status
DP container status affects the DP function.
UB->UBSVirt Info->Dp Privileged
true
UBSVirt Container Network-DP privileged
The DP privileged container affects the DP function.
UB->UBSE Info->UBSE Communication Mode
urma
UBSE-cluster.ipList
If the configuration is incorrect, URMA communication in the container may be unavailable.
UB->UBSE Info->Cert Use
FALSE
UBSE-cert.use
It indicates whether to use certificates for cross-node communication.
UB->UBSE Info->Start Status
Success
UBSE-UBSE status
If UBSE fails to start, the container startup and bonding delivery will fail.
UB->URMA Info->Urma Device Count
754
URMA-urma_admin show
If the device fails to be loaded or the loading is incomplete, the URMA bonding device cannot be aggregated, affecting basic functions.
bios->BIOS detail Info->Power Policy
Performance
URMA interrupt performance, UBS Comm-BIOS configuration
If the performance mode is not enabled, the interrupt packet sending performance deteriorates from the μs level to the ms level.
bios->BIOS detail Info->HiBoost
Enabled
URMA interrupt performance, UBS Comm-BIOS configuration
If the performance mode is not enabled, the interrupt packet sending performance deteriorates from the μs level to the ms level.
bios->BIOS detail Info->HiBoost Turbo+ Profile
Auto
URMA interrupt performance, UBS Comm-BIOS configuration
If the performance mode is not enabled, the interrupt packet sending performance deteriorates from the μs level to the ms level.
bios->BIOS detail Info->SMT2
Enabled
URMA interrupt performance, UBS Comm-BIOS configuration
If the performance mode is not enabled, the interrupt packet sending performance deteriorates from the μs level to the ms level.
memory->OS Memory Info->Transparent Huge Pages
never
Transparent huge pages
Enabling THP may trigger automatic page management in the system. When THP is set to always, 4 KB small pages are combined into 2 MB large pages. This process is transparent to users and applications and increases the TLB hit ratio. However, background CPU activity may increase, and the performance benefits are not always stable. Transparent huge pages cannot always be allocated as 2 MB pages, especially after the system runs for a long time and memory becomes fragmented.
memory->OS Memory Info->Automatic NUMA Balancing
Disabled
numa_balancing
If a service container is bound to a NUMA node and the node memory meets service requirements, it is recommended to disable this feature to avoid the overhead caused by background scanning.
UB->UBM Info->SlotId
[1,4]
UBM-slotid
The overall configuration of the management and control plane fails to be delivered.
UB->UBM Info->Scene
1d4n2p
UBM-scene
The overall configuration of the management and control plane fails to be delivered.
UB->UBM Info->MAMI Status
Success
UBM-MAMI running status
UBM startup is affected.
UB->UBM Info->LCNE UserId
\d+
UBM-User/User group
UBM startup is affected.
UB->UBM Info->LCNE Running Status
.*running.*
UBM-LCNE startup status
UB->UB BMC Info->IsUBFabricMode
TRUE
BMC-Cluster mode configuration
Whether the current node is in a UnifiedBus cluster affects UnifiedBus-based communication.
UB->UB BMC Info->IsUBControlNode
TRUE
BMC-Management node configuration
Whether the current node is a management or proxy node affects UnifiedBus-based communication.
UB->UB BMC Info->TopologyType
1
BMC-Networking mode configuration
The current networking has 1D and 2D modes, which affect UnifiedBus-based communication.
UB->UB BMC Info->UBMDeployMode
1
BMC-Management mode configuration
The current node is configured for in-band or out-of-band management, which affects UnifiedBus-based communication.
- Upload the prepared baseline file to the /home directory on the server.
- Generate a new collection report and compare it with the baseline report.
./kspect report --diff_base /home/base.xlsx
Command output:
Some data of the BMC, BIOS, storage, and UnifiedBus modules depends on the BMC login information. Are you sure you want to enter (y or N): Enter the BMC IP address: xx.xx.xx.xx BMC user name: Administrator BMC password: Collecting... Please wait. Collection completed. Total duration: 8.29 seconds The JSON file /home/devkit-kspect-x.x.x-Linux-aarch64/output/kspect-report-20260225-142117/kspect-json-20260225-142117.json has been generated. The CSV file /home/devkit-kspect-x.x.x-Linux-aarch64/output/kspect-report-20260225-142117/kspect-report-20260225-142117.csv has been generated. The HTML file /home/devkit-kspect-x.x.x-aarch64/output/kspect-report-20260225-142117/kspect-report-20260225-142117.html has been generated. Note: The [current] report is /home/devkit-kspect-x.x.x-Linux-aarch64/output/kspect-report-20260225-142117/kspect-json-20260225-142117.json. The [base] report is /home/base.xlsx. BIOS =========================================================================================================================================================== BIOS detail Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── LPI KV cache performance-CPU PM Contrl Disabled Disabled Consistent -- -- Power Policy URMA interrupt performance, UBS Comm-BIOS configuration Performance Performance Consistent -- -- HiBoost URMA interrupt performance, UBS Comm-BIOS configuration Enabled Enabled Consistent -- -- HiBoost Turbo+ Profile URMA interrupt performance, UBS Comm-BIOS configuration Auto Auto Consistent -- -- SMT2 URMA interrupt performance, UBS Comm-BIOS configuration Enabled Enabled Consistent -- -- ───────────────────────────────────────────────────────────────────────────── Memory ============================================================================================================================================================ OS Memory Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── Automatic NUMA Balancing numa_balancing Disabled Disabled Consistent -- -- Transparent Huge Pages Transparent pages never never Consistent -- -- ───────────────────────────────────────────────────────────────────────────── UB =========================================================================================================================================================== KVCache Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── Enable Urma KVCache Performance-enable_urma -- True Inconsistent The datasystem worker components pull data from each other through the URMA API. -- Enable Rdma KVCache Performance-enable_rdma -- True Inconsistent -- -- Urma Mode KVCache Performance-urma_mode -- UB Inconsistent The datasystem worker components pull data from each other through URMA and UnifiedBus, -- improving the performance of cross-node data pulling of KV cache. Enable Worker Worker Batch Get KVCache Performance-enable_worker_worker_batch_get -- True Inconsistent The datasystem worker components batch pull data from each other, -- improving the performance of cross-node data pulling of KV cache. ───────────────────────────────────────────────────────────────────────────── UBSE Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── UBSE Communication Mode UBSE-cluster.ipList urma urma Consistent -- -- Cert Use UBSE-cert.use false False Consistent -- -- Start Status UBSE-UBSE status Failed Success Inconsistent If UBSE fails to start, the container startup and bonding delivery will fail. -- ───────────────────────────────────────────────────────────────────────────── URMA Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── Urma Device Count URMA-urma_admin show -- 754 Inconsistent If the device fails to be loaded or the loading is incomplete, the URMA bonding device cannot be aggregated, -- affecting basic functions. ───────────────────────────────────────────────────────────────────────────── UBM Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── SlotId UBM-slotid 1 [1,4] Consistent -- -- Scene UBM-scene 1d8n2p 1d4n2p Inconsistent The overall configuration of the management and control plane fails to be delivered. -- MAMI Status UBM-MAMI running status Success Success Consistent -- -- LCNE UserId UBM-User/User Group 1000 \d+ Consistent -- -- LCNE Running Status UBM-LCNE startup status active (running) .*running.* Consistent -- -- ───────────────────────────────────────────────────────────────────────────── UB BMC Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── IsUBFabricMode BMC-Cluster mode configuration -- True Inconsistent Whether the current node is in a UnifiedBus cluster affects UnifiedBus-based communication. -- IsUBControlNode BMC-Management node configuration -- True Inconsistent Whether the current node is a management or proxy node affects UnifiedBus-based communication. -- TopologyType BMC-Networking mode configuration -- 1 Inconsistent The current networking has 1D and 2D modes, which affect UnifiedBus-based communication. -- UBMDeployMode BMC-Management mode configuration -- 1 Inconsistent The current node is configured for in-band or out-of-band management, which affects UnifiedBus-based communication. -- ───────────────────────────────────────────────────────────────────────────── UBSVirt Info ───────────────────────────────────────────────────────────────────────────── key name current base diff impact modification method ───────────────────────────────────────────────────────────────────────────── Pod Cpu Config UBSVirt Container Network-CPU Baseline 50m [50,]m Consistent -- -- Pod Memory Config UBSVirt Container Network-Memory Baseline 50Mi [50,]Mi Consistent -- -- Dp Pod Status UBSVirt Container Network-DP Container Status [Running]*2 running Inconsistent DP container status affects the DP function. -- Dp Privileged UBSVirt Container Network--DP privileged true true Consistent -- -- ───────────────────────────────────────────────────────────────────────────── The JSON file /home/devkit-kspect-x.x.x-Linux-aarch64/output/kspect-diff-20260225-142117/kspect-diff-20260225-142117.json has been generated.Baseline comparison supports regular expressions and range ([a,b]) matching. Units are supported for range matching. For example, [50,]Mib means greater than or equal to 50 Mib. If the diff value in the result is inconsistent, the impact and suggested modification are displayed.
- Modify the server based on the inconsistent items in the comparison report