Which plan provides Lowest latency to US located servers?

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
*post deleted*

Please ignore my post since I just realize your request is more for trading purposes than gaming.
Oh I saw your earlier post. You're right and I agree the risk still remains that even if today the ISP agrees to custom route there's no telling situation won't change down the road and I might be subjected at the ISP's mercy again when their routing change due to some upgrade or maintenance job or configuration change with their network and backend systems.

Thank you so much for sharing the alternative to go with Mudfish. I've not use Mudfish before though in the past I've used quite many VPNs i.e. expressvpn, nordvpn, protonvpn, vyprvpn, strongvpn, ipvanish, hma, windscribe. But for reducing latency in regards to trading purposes, I find them not much of a use.

Mudfish seemed really interesting and I found the ping tool at its website able to determine latency from its various nodes. And among the SG nodes, the best are Vultr and Google2 which are able to achieve 198ms-200ms latency to the IP addresses I highlighted earlier, and this latency is on par with SI's and VQ's.

May I check with you is Mudfish a korean company and if based on Mudfish pricing of 1GB international traffic = 135 won (krw), hypothetically let's say if I run up 500GB of traffic per month, I'll be looking at north of over SG$80? Or is there a better way to arrive at a rough idea of a likely overhead cost? Many thanks!
 

bert64

Senior Member
Joined
Jan 20, 2020
Messages
501
Reaction score
54
Or you could rent a vps from vultr directly for $2.50/month with 500gb of traffic, and run your own tunnel.
Do consider the latency from your own connection to vultr. as you would need to add that to the 190ms figure:
https://sgp-ping.vultr.com

From M1 i get around 4.3ms over ipv4 and 3.6ms over ipv6.

If you're using singtel then vultr won't work well as they don't peer locally, your traffic will go abroad and back to reach vultr's local nodes.
 

jasonho

Senior Member
Joined
Jan 1, 2000
Messages
992
Reaction score
17
Oh I saw your earlier post. You're right and I agree the risk still remains that even if today the ISP agrees to custom route there's no telling situation won't change down the road and I might be subjected at the ISP's mercy again when their routing change due to some upgrade or maintenance job or configuration change with their network and backend systems.

Thank you so much for sharing the alternative to go with Mudfish. I've not use Mudfish before though in the past I've used quite many VPNs i.e. expressvpn, nordvpn, protonvpn, vyprvpn, strongvpn, ipvanish, hma, windscribe. But for reducing latency in regards to trading purposes, I find them not much of a use.

Mudfish seemed really interesting and I found the ping tool at its website able to determine latency from its various nodes. And among the SG nodes, the best are Vultr and Google2 which are able to achieve 198ms-200ms latency to the IP addresses I highlighted earlier, and this latency is on par with SI's and VQ's.

May I check with you is Mudfish a korean company and if based on Mudfish pricing of 1GB international traffic = 135 won (krw), hypothetically let's say if I run up 500GB of traffic per month, I'll be looking at north of over SG$80? Or is there a better way to arrive at a rough idea of a likely overhead cost? Many thanks!
Yes, its a korean company. I feel it all depends on your needs, I will cater generally $50-100 for heavy bandwidth and $20 for gaming (not 24/7 of course). It costs me less than $25 each month for WoW and other games.

Adv of mudfish is latency and flexible routing connection. Like I mentioned before, even a good routing now may change because there's so much variables (or connection points) between you and the final destination.

Another more expensive (& troublesome) option is like what @bert64 suggested. You can try vultr or even GCP (Google Cloud Platform) but the cons is you have to tinkle with setting up VPS and routing (software setup). I know friends who get 900+Mbps speed connecting to US using speedtest. GCP was offering USD$300 credits and you can try setup multiple GCP vps and try pinging each one to determine the best route.

Nevertheless, given what your objective is, you may or may not want to spend too much time on VPS options. The more connection points there are, the more points of (possible) failures. So you have to decide whats the best option :)
 
Last edited:

bert64

Senior Member
Joined
Jan 20, 2020
Messages
501
Reaction score
54
unfortunately due to singapore laws, private ownership of any satellite dishes are prohibited. how to starlink :(
Learn something every day...
In a lot of countries, satellite tv systems are common so lots of people have satellite dishes bolted to the side of their homes.
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
Or you could rent a vps from vultr directly for $2.50/month with 500gb of traffic, and run your own tunnel.
Do consider the latency from your own connection to vultr. as you would need to add that to the 190ms figure:
https://sgp-ping.vultr.com

From M1 i get around 4.3ms over ipv4 and 3.6ms over ipv6.

If you're using singtel then vultr won't work well as they don't peer locally, your traffic will go abroad and back to reach vultr's local nodes.
Yes, its a korean company. I feel it all depends on your needs, I will cater generally $50-100 for heavy bandwidth and $20 for gaming (not 24/7 of course). It costs me less than $25 each month for WoW and other games.

Adv of mudfish is latency and flexible routing connection. Like I mentioned before, even a good routing now may change because there's so much variables (or connection points) between you and the final destination.

Another more expensive (& troublesome) option is like what @bert64 suggested. You can try vultr or even GCP (Google Cloud Platform) but the cons is you have to tinkle with setting up VPS and routing (software setup). I know friends who get 900+Mbps speed connecting to US using speedtest. GCP was offering USD$300 credits and you can try setup multiple GCP vps and try pinging each one to determine the best route.

Nevertheless, given what your objective is, you may or may not want to spend too much time on VPS options. The more connection points there are, the more points of (possible) failures. So you have to decide whats the best option :)

Nice, thanks for suggesting that, gentlemen. This sounds like a workable idea. I have had some experience setting up own vpn server with AWS using openvpn protocol in the past but that was like ten years ago in my previous live when I was based out of china trying to bypass their censorship. And time flies but I've not caught up with this part of the technology evolution for a long time. @bert64 If you have come across any spoon feeding guides out there which caters to Vultr and/or GCP do share them with me. Many thanks in advance.

A brief googling and I found these two articles. Hopefully still relevant in today's context. :D
https://mydoto.com/google-cloud-platform/how-to-create-your-own-vpn-in-google-cloud-platform/
https://newbloghosting.com/how-to-create-your-own-vpn-in-vultr/

@jasonho Indeed, if its not quick and little effort, I might just fall back to Mudfish. That nerdy side of me has been awoken so got to just try it out and get my hands dirty first. Let's see.

M1 came back today and said 235ms is the best they get. Can't further reduce the number of hops or latency anymore. :cry: Oh well...
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
Wanted to provide an update on the progress with my effort to reduce latency. In case anyone faces similar needs as me, know that these options have been explored.

Further liaison with SI ends up nowhere. According to them, they won't help customers facing latency issue. So SI is now strike off my sign-up list. Same case with MR, which refuses outright to assist in cases of such nature. So in the end I am left with VQ whose sales side is willing to help but it turns out VQ's actual traceroute test with one of the needed IPs is even worst then M1 at more than 250ms.
|------------------------------------------------------------------------------------------|
| WinMTR statistics |
| Host - % | Sent | Recv | Best | Avrg | Wrst | Last |
|------------------------------------------------|------|------|------|------|------|------|
| router - 0 | 143 | 143 | 0 | 1 | 20 | 2 |
| VQsg - 0 | 143 | 143 | 2 | 6 | 66 | 2 |
| VQ LA - 0 | 143 | 143 | 184 | 187 | 246 | 184 |
|te0-1-0-16.ccr41.lax04.atlas.cogentco.com - 0 | 143 | 143 | 179 | 180 | 196 | 179 |
| be3360.ccr42.lax01.atlas.cogentco.com - 0 | 143 | 143 | 179 | 181 | 198 | 179 |
| be2932.ccr32.phx01.atlas.cogentco.com - 0 | 143 | 143 | 190 | 192 | 210 | 191 |
| be2930.ccr21.elp01.atlas.cogentco.com - 0 | 143 | 143 | 202 | 203 | 221 | 202 |
| be2927.ccr41.iah01.atlas.cogentco.com - 0 | 143 | 143 | 208 | 210 | 226 | 210 |
| be2687.ccr41.atl01.atlas.cogentco.com - 0 | 143 | 143 | 223 | 226 | 272 | 223 |
| be3364.rcr21.ind01.atlas.cogentco.com - 0 | 143 | 143 | 239 | 241 | 260 | 239 |
| be2721.rcr51.cvg02.atlas.cogentco.com - 0 | 143 | 143 | 236 | 238 | 271 | 236 |
|be2600.nr01.b038762-0.cvg02.atlas.cogentco.com - 0 | 143 | 143 | 236 | 238 | 256 | 237 |
| 38.122.234.18 - 0 | 143 | 143 | 242 | 243 | 259 | 242 |
| be-4003-r0.chi2.cyrusone.net - 0 | 143 | 143 | 244 | 245 | 260 | 244 |
| 216.117.22.150 - 0 | 143 | 143 | 246 | 247 | 264 | 247 |
| 64.208.51.207 - 0 | 143 | 143 | 244 | 246 | 259 | 244 |
|________________________________________________|______|______|______|______|______|______|
WinMTR v0.92 GPL V2 by Appnor MSP - Fully Managed Hosting & Cloud Provider

Therefore it seemed possibility remains that the previous Ripe Atlas tests with result of around 200ms might not be genuine VQ users after all.

One of the previous advice in the earlier posts indicated that all ISPs are the same, and perhaps to a large extend this is true. So my effort on pursuing custom routes option with ISPs shall end here.
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
As to other options with using Mudfish VPN and setting up own tunnel, there has been some progress and hurdle.

On Mudfish VPN, I have been able to cut down latency to around 200ms via a combination of Google 2 and Vultr nodes. But it seemed one step forward is met with another step backward. Though latency is cut down but throughput became exceptionally bad. With VPN, the top speed obtainable is only 1/3 of normal speed. Not sure how to improve on this issue yet.

As for setting up own tunnel, I managed to setup 3 openvpn servers, 3 ipsec servers, and 3 wireguard servers with GCP for each of the 3 zones in Singapore. By using GCP's premium network, latency were also cut down to only 8 hops and average 200ms, which is probably the best that can be obtain with current undersea cables systems. However, the same issue was faced with throughput as well.

For openvpn, speed (via fast.com tests) is reduced to only 1/5 that of normal speed (where vpn was not turn on). Whereas for ipsec, it is slightly better at 1/4 speed. Wireguard give the best speed but still limited to 1/3 that of normal.

Anyone know how to improve throughput in such scenario? Not sure if it is my current ISP (M1) throttling vpn or is the slowness caused by my Asus router's firewall. Will need to do more research on this issue.
 

xiaofan

Supremacy Member
Joined
Sep 16, 2018
Messages
9,477
Reaction score
71
as for vpn speed, why not try installing v2ray instead? they have good throughput

V2ray is not bad, or Trojan. I have not tried newer ones like x-ray yet. Since I am using free Oracle cloud (US servers) the speed is not great though.
 

bert64

Senior Member
Joined
Jan 20, 2020
Messages
501
Reaction score
54
As to other options with using Mudfish VPN and setting up own tunnel, there has been some progress and hurdle.

On Mudfish VPN, I have been able to cut down latency to around 200ms via a combination of Google 2 and Vultr nodes. But it seemed one step forward is met with another step backward. Though latency is cut down but throughput became exceptionally bad. With VPN, the top speed obtainable is only 1/3 of normal speed. Not sure how to improve on this issue yet.

As for setting up own tunnel, I managed to setup 3 openvpn servers, 3 ipsec servers, and 3 wireguard servers with GCP for each of the 3 zones in Singapore. By using GCP's premium network, latency were also cut down to only 8 hops and average 200ms, which is probably the best that can be obtain with current undersea cables systems. However, the same issue was faced with throughput as well.

For openvpn, speed (via fast.com tests) is reduced to only 1/5 that of normal speed (where vpn was not turn on). Whereas for ipsec, it is slightly better at 1/4 speed. Wireguard give the best speed but still limited to 1/3 that of normal.

Anyone know how to improve throughput in such scenario? Not sure if it is my current ISP (M1) throttling vpn or is the slowness caused by my Asus router's firewall. Will need to do more research on this issue.
If you are solely concerned with latency/throughput and not about encryption, perhaps use SIT/GRE tunnels as they would incur less overhead.
Also be mindful that any form of tunnelling will decrease the MTU which will slightly hurt throughput.

In terms of speed reduction, what are you comparing? testing downloading of something from your target server(s)? If your testing local vs further away then throughput will be reduced due to the higher latency, you can play with tcp window scaling options etc but also be aware that for a small transfer it won't have time to ramp up.
 

bert64

Senior Member
Joined
Jan 20, 2020
Messages
501
Reaction score
54
Wanted to provide an update on the progress with my effort to reduce latency. In case anyone faces similar needs as me, know that these options have been explored.

Further liaison with SI ends up nowhere. According to them, they won't help customers facing latency issue. So SI is now strike off my sign-up list. Same case with MR, which refuses outright to assist in cases of such nature. So in the end I am left with VQ whose sales side is willing to help but it turns out VQ's actual traceroute test with one of the needed IPs is even worst then M1 at more than 250ms.
|------------------------------------------------------------------------------------------|
| WinMTR statistics |
| Host - % | Sent | Recv | Best | Avrg | Wrst | Last |
|------------------------------------------------|------|------|------|------|------|------|
| router - 0 | 143 | 143 | 0 | 1 | 20 | 2 |
| VQsg - 0 | 143 | 143 | 2 | 6 | 66 | 2 |
| VQ LA - 0 | 143 | 143 | 184 | 187 | 246 | 184 |
|te0-1-0-16.ccr41.lax04.atlas.cogentco.com - 0 | 143 | 143 | 179 | 180 | 196 | 179 |
| be3360.ccr42.lax01.atlas.cogentco.com - 0 | 143 | 143 | 179 | 181 | 198 | 179 |
| be2932.ccr32.phx01.atlas.cogentco.com - 0 | 143 | 143 | 190 | 192 | 210 | 191 |
| be2930.ccr21.elp01.atlas.cogentco.com - 0 | 143 | 143 | 202 | 203 | 221 | 202 |
| be2927.ccr41.iah01.atlas.cogentco.com - 0 | 143 | 143 | 208 | 210 | 226 | 210 |
| be2687.ccr41.atl01.atlas.cogentco.com - 0 | 143 | 143 | 223 | 226 | 272 | 223 |
| be3364.rcr21.ind01.atlas.cogentco.com - 0 | 143 | 143 | 239 | 241 | 260 | 239 |
| be2721.rcr51.cvg02.atlas.cogentco.com - 0 | 143 | 143 | 236 | 238 | 271 | 236 |
|be2600.nr01.b038762-0.cvg02.atlas.cogentco.com - 0 | 143 | 143 | 236 | 238 | 256 | 237 |
| 38.122.234.18 - 0 | 143 | 143 | 242 | 243 | 259 | 242 |
| be-4003-r0.chi2.cyrusone.net - 0 | 143 | 143 | 244 | 245 | 260 | 244 |
| 216.117.22.150 - 0 | 143 | 143 | 246 | 247 | 264 | 247 |
| 64.208.51.207 - 0 | 143 | 143 | 244 | 246 | 259 | 244 |
|________________________________________________|______|______|______|______|______|______|
WinMTR v0.92 GPL V2 by Appnor MSP - Fully Managed Hosting & Cloud Provider

Therefore it seemed possibility remains that the previous Ripe Atlas tests with result of around 200ms might not be genuine VQ users after all.

One of the previous advice in the earlier posts indicated that all ISPs are the same, and perhaps to a large extend this is true. So my effort on pursuing custom routes option with ISPs shall end here.
For the atlas tests i think i only tested to one of your destinations...
I'm surprised SI aren't willing to look at routing as i thought they were a more premium provider, have been quite impressed with M1 as they are actually willing to look into it, albeit with no success in your instance.

Is what your doing actually throughput sensitive, or only latency? You could just split the route - latency sensitive servers through vpn, other traffic direct etc.

setting up openvpn on vultr, gcp, whatever will be exactly the same as aws as it's all the same linux distros you'll install in the vm..

BTW if you're hosting a vpn endpoint at azure/aws/vultr you can shave 1-2ms off by doing the first link M1->vultr over ipv6.
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
as for vpn speed, why not try installing v2ray instead? they have good throughput
V2ray is not bad, or Trojan. I have not tried newer ones like x-ray yet. Since I am using free Oracle cloud (US servers) the speed is not great though.

Thank you both for suggesting these options. I did some reading up and indeed it seemed Xray gives the best throughput, followed by Trojan (or the newer Trojan-go), followed by V2ray.

However it seemed these are created by the chinese community primarily aimed to bypass GFW censorship and hence uses protocols such as shadowsocks or socks5 proxies to mask IP and traffic instead of actually creating tunnel to route traffic?

Actually, my requirements are 2 folds, to (i) reduce latency while (ii) maintaining as high a throughput as possible. But it seemed the more ancient VPN tunneling methods like openvpn, ipsec and wireguard can only satisfy the first requirement while newer options (xray, trojan, v2ray) can only address the latter. Seemed like there is no best of both worlds. Or am I mistaken?
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
as for vpn speed, why not try installing v2ray instead? they have good throughput
V2ray is not bad, or Trojan. I have not tried newer ones like x-ray yet. Since I am using free Oracle cloud (US servers) the speed is not great though.

Just wanna say that I am not technical. More like a layman who can only follow step by step instructions. So I tried over 10 tutorials (refer links below) out there that I can find but unfortunately all efforts ended up failing miserably...😓




While I was trying to dive deeper to make them work, I realized the client-side software (for Windows) showed malware detection on virus total, for example v2rayN-Core, V2RayW. Might be false positives but as I'm using my machine for trading, I'd better don't take the chance.
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
If you are solely concerned with latency/throughput and not about encryption, perhaps use SIT/GRE tunnels as they would incur less overhead.
Also be mindful that any form of tunnelling will decrease the MTU which will slightly hurt throughput.

In terms of speed reduction, what are you comparing? testing downloading of something from your target server(s)? If your testing local vs further away then throughput will be reduced due to the higher latency, you can play with tcp window scaling options etc but also be aware that for a small transfer it won't have time to ramp up.

Many thanks. Do you know how to setup SIT/GRE tunnels? Any guide would be appreciated!

Tried doing some research on them but the only relevant info I can find is Google cloud don't support GRE.

For speed, I am testing from my machine using fast.com using each of the VPN options (openvpn, ipsec, wireguard) vs turning off vpn.

Come to think of it, its really true that small transfer don't have time to ramp up. And given the nature of my needs surrounding financial market trading where entering trade and exiting trade in matter of a few seconds, each trade's order (in terms of data size) which gets submitted to the exchange servers in Chicago for each and every trade would be really small (few bytes to few KBs perhaps...) so maybe throughput won't matter that much after all?...hmmmm...
 

adderrs

Member
Joined
Dec 19, 2016
Messages
174
Reaction score
4
Is what your doing actually throughput sensitive, or only latency? You could just split the route - latency sensitive servers through vpn, other traffic direct etc.

setting up openvpn on vultr, gcp, whatever will be exactly the same as aws as it's all the same linux distros you'll install in the vm..

BTW if you're hosting a vpn endpoint at azure/aws/vultr you can shave 1-2ms off by doing the first link M1->vultr over ipv6.

You asked a very good question. 😀 Come to think of it might not be throughput sensitive (per my illustration above) but I really can't tell for now nor have any tool or idea how to differentiate as nothing much can be observe from Win10 resource monitor except the usual stuff like remote addresses, ports, etc.

I'm currently testing those VPN options on GCP only, since Azure and AWS don't provide latency in the 200-205ms range from the earlier Ripe atlas result and mudfish traceroute which uses the same nodes. Had also tested Vultr but it turn out that latency is no better than M1's for one of the needed IP address, while GCP gives the lowest latency for both IPs.

May I ask how to enable ipv6 from M1 to GCP? Is there a way to force ipv6 for that first leg? Only way I can get ipv6 to work is when torrenting. lol... 😁
 

miloaisdino

Senior Member
Joined
Mar 25, 2016
Messages
878
Reaction score
19
You asked a very good question. 😀 Come to think of it might not be throughput sensitive (per my illustration above) but I really can't tell for now nor have any tool or idea how to differentiate as nothing much can be observe from Win10 resource monitor except the usual stuff like remote addresses, ports, etc.

I'm currently testing those VPN options on GCP only, since Azure and AWS don't provide latency in the 200-205ms range from the earlier Ripe atlas result and mudfish traceroute which uses the same nodes. Had also tested Vultr but it turn out that latency is no better than M1's for one of the needed IP address, while GCP gives the lowest latency for both IPs.

May I ask how to enable ipv6 from M1 to GCP? Is there a way to force ipv6 for that first leg? Only way I can get ipv6 to work is when torrenting. lol... 😁
ipv4 from m1 to gcp should be quite low already (single digit). imo theres no need for ipv6
 
Important Forum Advisory Note
This forum is moderated by volunteer moderators who will react only to members' feedback on posts. Moderators are not employees or representatives of HWZ. Forum members and moderators are responsible for their own posts.

Please refer to our Terms of Service for more information.
Top