[GH-ISSUE #4382] Paralel requests performance

gitea-mirror commented

2026-05-05 14:13:36 -06:00

Owner

Originally created by @blue-genie on GitHub (Aug 15, 2024).
Original GitHub issue: https://github.com/fatedier/frp/issues/4382

Bug Description

This is not really a bug, but I would like to get the community's recommendation on how to implement for my use case.

I have a web server for which I have 2 ISPs, one primary and a 2nd as backup for when the primary goes down, which runs frpc.
I also have a AWS server which runs frps and handles XXX.ZZZ.com and YYY.ZZZ.com, which is supposed to forward the requests to the in-house server.
The configuration for the client and server are below.
I have a page which on initial load, loads maybe about 30MBs, in 20-25 images, the images are supposed to be loaded in parallel.

frpc Version

fatedier/frps:v0.59.0

frps Version

fatedier/frps:v0.59.0

System Architecture

linux/amd64 all server

Configurations

# frps.toml
bindPort = 7000
vhostHTTPPort = 80
vhostHTTPSPort = 9443

# frpc.toml
serverAddr = "XXX.XXX.XXX.XXX"
serverPort = 7000

[[proxies]]
name = "XXXX-https"
#type = "tcp"
type = "https"
localIP = "nginx"
localPort = 443
customDomains = ["XXX.ZZZ.com", "YYY.ZZZ.com"]

# now v1 and v2 are supported
transport.proxyProtocolVersion = "v2"

[[proxies]]
name = "XXX-http"
type = "http"
localIP = "nginx"
localPort = 80 
customDomains = ["XXX.ZZZ.com", "YYY.ZZZ.com"]

Logs

no errors just optimization query

Steps to reproduce

Issue, using the same client computer - just messing with the posts file to decide if the request is sent to AWS&frps or directly to the server:

If my clients go directly to the server, bypassing AWS & frps, they get really fast, 3x, on a full page load, with cache disabled and 33 http requests.

If the request goes through frps I see performance issues. The page that used to load in 3s not it's loaded in 9-10s. My nginx server reports that the http request time, $request_time, was 0.5s, but the some images load in 6s - 3-4s waiting for server response and 2-3s content download

Question: Is my configuration optimum for my needs? Anything that I should change?

Affected area

Docs
Installation
Performance and Scalability
Security
User Experience
Test and Release
Developer Infrastructure
Client Plugin
Server Plugin
Extensions
Others

Originally created by @blue-genie on GitHub (Aug 15, 2024). Original GitHub issue: https://github.com/fatedier/frp/issues/4382 ### Bug Description This is not really a bug, but I would like to get the community's recommendation on how to implement for my use case. I have a web server for which I have 2 ISPs, one primary and a 2nd as backup for when the primary goes down, which runs frpc. I also have a AWS server which runs frps and handles XXX.ZZZ.com and YYY.ZZZ.com, which is supposed to forward the requests to the in-house server. The configuration for the client and server are below. I have a page which on initial load, loads maybe about 30MBs, in 20-25 images, the images are supposed to be loaded in parallel. ### frpc Version fatedier/frps:v0.59.0 ### frps Version fatedier/frps:v0.59.0 ### System Architecture linux/amd64 all server ### Configurations ``` # frps.toml bindPort = 7000 vhostHTTPPort = 80 vhostHTTPSPort = 9443 ``` ``` # frpc.toml serverAddr = "XXX.XXX.XXX.XXX" serverPort = 7000 [[proxies]] name = "XXXX-https" #type = "tcp" type = "https" localIP = "nginx" localPort = 443 customDomains = ["XXX.ZZZ.com", "YYY.ZZZ.com"] # now v1 and v2 are supported transport.proxyProtocolVersion = "v2" [[proxies]] name = "XXX-http" type = "http" localIP = "nginx" localPort = 80 customDomains = ["XXX.ZZZ.com", "YYY.ZZZ.com"] ``` ### Logs no errors just optimization query ### Steps to reproduce Issue, using the same client computer - just messing with the posts file to decide if the request is sent to AWS&frps or directly to the server: If my clients go directly to the server, bypassing AWS & frps, they get really fast, 3x, on a full page load, with cache disabled and 33 http requests. If the request goes through frps I see performance issues. The page that used to load in 3s not it's loaded in 9-10s. My nginx server reports that the http request time, `$request_time`, was 0.5s, but the some images load in 6s - 3-4s `waiting for server response` and 2-3s `content download` Question: Is my configuration optimum for my needs? Anything that I should change? ### Affected area - [ ] Docs - [ ] Installation - [X] Performance and Scalability - [ ] Security - [ ] User Experience - [ ] Test and Release - [ ] Developer Infrastructure - [ ] Client Plugin - [ ] Server Plugin - [ ] Extensions - [ ] Others

gitea-mirror

2026-05-05 14:13:36 -06:00

closed this issue
added the
lifecycle/stale
label

gitea-mirror commented

2026-05-05 14:13:39 -06:00

Author

Owner

@fatedier commented on GitHub (Aug 15, 2024):

This usually depends on your network conditions and bandwidth.

@fatedier commented on GitHub (Aug 15, 2024): This usually depends on your network conditions and bandwidth.

gitea-mirror commented

2026-05-05 14:13:40 -06:00

Author

Owner

@blue-genie commented on GitHub (Aug 15, 2024):

In general I would agree, but I would assume that AWS has symmetric 1Gbps connection. I have 1Gbps symmetric on the server.

I understand that there will be an extra delay, because the request goes from local to AWS then comes back a different local, instead just between two locals, but still ...

Is my configuration the most optimum for the high concurent traffic? I tried to mess with the connection pooling, didn't see a difference.

What would be your configuration for this setup? Or how do I get more stats to debug and see if anything can be optimized?

@blue-genie commented on GitHub (Aug 15, 2024): In general I would agree, but I would assume that AWS has symmetric 1Gbps connection. I have 1Gbps symmetric on the server. I understand that there will be an extra delay, because the request goes from local to AWS then comes back a different local, instead just between two locals, but still ... Is my configuration the most optimum for the high concurent traffic? I tried to mess with the connection pooling, didn't see a difference. What would be your configuration for this setup? Or how do I get more stats to debug and see if anything can be optimized?

gitea-mirror commented

2026-05-05 14:13:41 -06:00

Author

Owner

@github-actions[bot] commented on GitHub (Sep 6, 2024):

Issues go stale after 21d of inactivity. Stale issues rot after an additional 7d of inactivity and eventually close.

@github-actions[bot] commented on GitHub (Sep 6, 2024): Issues go stale after 21d of inactivity. Stale issues rot after an additional 7d of inactivity and eventually close.

gitea-mirror referenced this issue

2026-05-05 14:48:04 -06:00

[PR #3460] [MERGED] xtcp: when connection timeout occurs, support fallback to STCP #4753

gitea-mirror referenced this issue

2026-05-05 14:48:15 -06:00

[PR #3499] [MERGED] release v0.50.0 #4763