秋田预测市场 · Discord 审计

订单创建时并行请求与批处理请求的延迟比较

2026-06-21 · Kalshi · 5 条相关讨论

olimano1 2026-06-21 00:29:58

are people actually seeing parallel single order creation be lower latency than batch? I've been experimenting ahead of transitioning to single order, and am fairly consistently finding the batch to be marginally lower latency than median parallel single order

are people actually seeing parallel single order creation be lower latency than batch? I've been experimenting ahead of transitioning to single order, and am fairly consistently finding the batch to be marginally lower latency than median parallel single order

Bumbling Bayesian 2026-06-21 00:34:52

I tested various approaches using the CloudFront hostname a while back and found it could be faster in some cases to use the batching.

I tested various approaches using the CloudFront hostname a while back and found it could be faster in some cases to use the batching.

Bumbling Bayesian 2026-06-21 00:36:19

I think it really depends on your use case. If you can have multiple requests in flight in parallel and don’t wait for the response, it can be faster to take advantage of the load balancer.

I think it really depends on your use case. If you can have multiple requests in flight in parallel and don’t wait for the response, it can be faster to take advantage of the load balancer.

Bumbling Bayesian 2026-06-21 00:38:12

But if your requests are for the same market, they are probably going to be serialized at some point anyway.

But if your requests are for the same market, they are probably going to be serialized at some point anyway.

olimano1 2026-06-21 00:58:20

hmm yes, seems like I don't have enough threads to fully parallelize signing 20 orders, so batches of 20 is more efficient because of that bottleneck

hmm yes, seems like I don't have enough threads to fully parallelize signing 20 orders, so batches of 20 is more efficient because of that bottleneck