SQD Worker Node Benchmark

It's the tool to determine hardware worthiness for running SQD worker. This tool runs number of checks to determine number of CPU cores, total amount of available memory, storage speed, available space, network and processing speed. Also it will try to determine your public IP.

The best way to run benchmark is to just use docker.

If you only have one storage device, you can run:

docker run --rm --network host --volume -it subsquid/worker-benchmark:latest

If you have dedicated storage for worker, please provide docker access to the expected storage:

docker run --rm --network host --volume [path to storage folder]:/app/data -it subsquid/worker-benchmark:latest

Expected output contains progress indicators, detailed performance breakdown and overall report.

Please note that as bulk of required bandwidth used to download initial data from infrastructure, we are measuring exactly this download speed. It may show numbers less than expected and usually not a dealbreaker as long as stated network speed is more or equal 1 Gigabit.

If tool is unable to detect you public IP, you would need to add it to worker configuration before running.

Blue means that parameter exceeding expectations and future-proof. Green means just right. Red in numbers other than network speed may lead to reduced APR.

If you are feeling generous, please submit "benchmark details" line along with your hardware spec, it will help further refine worker requirements.

Detailed run arguments

If you want to mess with internals, here are more details. If simply build and run, tool would do most of the data processing in ./data folder. Additional configuration includes:

• --total-samples: total number of queries to run, more makes benchmarking more precise, less makes benchmark take less time. Default is 1000, change would invalidate bench time in report.
• --max-concurrency: max number of queries to be run in a batch. As backend already take advantage of parallelization, it's not really useful for current implementation on machines with less than 16 cores. Default is 1, change would invalidate bench time in report.
• --config-url: URL to fetch benchmark description.
• --config-file: filename to save benchmark description.
• --force-update: re-download benchmark data even it is already downloaded.
• --no-cleanup: disables auto-cleanup.

1000 samples

Sol #1 query as example:

{
"type":"solana",
"fields":{
	"block":{"number":true,"hash":true,"parentNumber":true,"parentHash":true,"height":true,"timestamp":true},
	"transaction":{"transactionIndex":true,"accountKeys":true,"numReadonlySignedAccounts":true,"numReadonlyUnsignedAccounts":true,"numRequiredSignatures":true,"signatures":true,"fee":true,"computeUnitsConsumed":true,"loadedAddresses":true,"hasDroppedLogMessages":true},
	"instruction":{"transactionIndex":true,"instructionAddress":true,"programId":true,"accounts":true,"data":true,"d1":true,"d2":true,"d4":true,"d8":true,"error":true,"computeUnitsConsumed":true,"isCommitted":true,"hasDroppedLogMessages":true},
	"log":{"transactionIndex":true,"logIndex":true,"instructionAddress":true,"programId":true,"kind":true,"message":true},
	"balance":{"account":true,"pre":true,"post":true},
	"tokenBalance":{"transactionIndex":true,"account":true,"preMint":true,"postMint":true,"preDecimals":true,"postDecimals":true,"preProgramId":true,"postProgramId":true,"preOwner":true,"postOwner":true,"preAmount":true,"postAmount":true},
	"reward":{"pubkey":true,"lamports":true,"postBalance":true,"rewardType":true,"commission":true}
},
"includeAllBlocks":true,
"instructions":[{
	"programId":["TokenkegQfeZyiNwAJbNbGKPFXCWuBvf9Ss623VQ5DA","TokenzQdBNbLqP5VEhdkAS6EPFLC1PHnBqCXEpPxuEb"],
	"transaction":true,
	"transactionBalances":true,
	"transactionTokenBalances":true,
	"transactionInstructions":true,
	"innerInstructions":true,
	"logs":true
}]}

Leaderboard

1 thread

Spec	Sol #1	Sol #2	Sol #3	Sol #4	Eth #1	Eth #2	Eth #3	Eth #4	Base	Arbitrum	Total (s)
Slowest 10%	512 ms	236 ms	227 ms	135 ms	530 ms	217 ms	114 ms	46 ms	23 ms	32 ms	~207 s
GCP E2 2vcpu	484 ms	189 ms	168 ms	97 ms	416 ms	151 ms	101 ms	29 ms	10 ms	27 ms	169 s
Slowest 25%	381 ms	194 ms	153 ms	65 ms	379 ms	185 ms	75 ms	38 ms	12 ms	21 ms	~150 s
Servarica 4vcpu	386 ms	132 ms	119 ms	70 ms	360 ms	155 ms	73 ms	27 ms	11 ms	22 ms	138 s
GCP E2 4vcpu	345 ms	115 ms	101 ms	57 ms	294 ms	110 ms	57 ms	20 ms	8 ms	17 ms	114 s
GCP N2 4vcpu	260 ms	86 ms	77 ms	43 ms	223 ms	83 ms	43 ms	15 ms	6 ms	12 ms	86 s
Median	197 ms	90 ms	83 ms	49 ms	200 ms	102 ms	46 ms	28 ms	6 ms	15 ms	~82 s
GCP E2 8vcpu	191 ms	61 ms	53 ms	28 ms	180 ms	72 ms	26 ms	13 ms	6 ms	9 ms	65 s
GCP E2 16vcpu	181 ms	55 ms	47 ms	25 ms	177 ms	75 ms	22 ms	13 ms	6 ms	9 ms	63 s
GCP N2 16vcpu	193 ms	50 ms	42 ms	22 ms	177 ms	70 ms	20 ms	11 ms	6 ms	8 ms	61 s
Fastest 10%	167 ms	49 ms	45 ms	21 ms	175 ms	75 ms	25 ms	12 ms	4 ms	12 ms	~59 s
i7-9700K 8 cores	152 ms	43 ms	37 ms	20 ms	138 ms	52 ms	27 ms	10 ms	4 ms	9 ms	50 s
M3 Max 16 cores	126 ms	49 ms	45 ms	31 ms	121 ms	51 ms	22 ms	15 ms	4 ms	6 ms	47 s
GCP C4 16vcpu	117 ms	28 ms	23 ms	12 ms	112 ms	42 ms	11 ms	6 ms	3 ms	4 ms	36 s
Velia Baremetal: 9950 32 cores	89 ms	22 ms	18 ms	10 ms	82 ms	32 ms	10 ms	5 ms	2 ms	4 ms	28 s

10 threads (Experimental)

Spec	Sol #1	Sol #2	Sol #3	Sol #4	Eth #1	Eth #2	Eth #3	Eth #4	Base	Arbitrum	Total (s)
GCP E2 2vcpu	1404 ms	341 ms	318 ms	528 ms	694 ms	255 ms	166 ms	57 ms	13 ms	28 ms	153 s
GCP E2 4vcpu	719 ms	612 ms	524 ms	306 ms	506 ms	302 ms	160 ms	52 ms	15 ms	30 ms	86 s
GCP N2 4vcpu	545 ms	514 ms	463 ms	284 ms	407 ms	232 ms	122 ms	39 ms	12 ms	24 ms	67 s
GCP E2 8vcpu	295 ms	232 ms	213 ms	166 ms	319 ms	203 ms	185 ms	41 ms	13 ms	30 ms	34 s
i7-9700K 8 cores	278 ms	225 ms	216 ms	169 ms	273 ms	212 ms	181 ms	40 ms	13 ms	27 ms	30 s
GCP E2 16vcpu	252 ms	196 ms	181 ms	145 ms	260 ms	188 ms	131 ms	42 ms	14 ms	43 ms	27 s
GCP N2 16vcpu	269 ms	193 ms	184 ms	138 ms	259 ms	183 ms	117 ms	41 ms	22 ms	45 ms	28 s
M3 Max 16 cores	195 ms	179 ms	166 ms	137 ms	208 ms	144 ms	132 ms	58 ms	9 ms	20 ms	23 s
GCP C4 16vcpu	156 ms	113 ms	101 ms	78 ms	159 ms	107 ms	62 ms	24 ms	10 ms	16 ms	16 s