My next motherboard

[JackR]

Your local area network (LAN) is not your problem, the wide area network (WAN) is.

The controlling/predominant factor is the Internet latency. The 180 millisecond time is probably a roundtrip (from your computer to your broker's server and back to your computer). That indicates about a 90 millisecond one-way path. If you are receiving streaming quotes from your broker then the telecom path (and associated routers) delay is slightly less than one-tenth of a second (100 milliseconds). It is difficult to know what your software does but your hardware probably adds less than 1 millisecond (100 microseconds)to that time. The only way to figure it out is by benchmarking your computer.

As you can see, if you are autotrading and you have a program processing ticks and then sending a buy, sell, or cancel based on that data, gaining a few microseconds would mean nothing compared to the path (Internet latency) delay you have vs someone relatively local to the server. You may be electronically closer to the server than someone in California.

 

[nitro]

I have nothing to add.

nitro

Quote from JackR:

Your local area network (LAN) is not your problem, the wide area network (WAN) is.

The controlling/predominant factor is the Internet latency. The 180 millisecond time is probably a roundtrip (from your computer to your broker's server and back to your computer). That indicates about a 90 millisecond one-way path. If you are receiving streaming quotes from your broker then the telecom path (and associated routers) delay is slightly less than one-tenth of a second (100 milliseconds). It is difficult to know what your software does but your hardware probably adds less than 1 millisecond (100 microseconds)to that time. The only way to figure it out is by benchmarking your computer.

As you can see, if you are autotrading and you have a program processing ticks and then sending a buy, sell, or cancel based on that data, gaining a few microseconds would mean nothing compared to the path (Internet latency) delay you have vs someone relatively local to the server. You may be electronically closer to the server than someone in California.

 

[Bernard111]

Quote from JackR:

The only way to figure it out is by benchmarking your computer.

Yes, 180ms is roundtrip.
I understand that essentially I cannot do anything for WAN but
which test should I run exactly on my PC?

Another question: would a VPN connection beneficial for speed or is just useful to skip the 15 hops?

 

[nitro]

Quote from Bernard111:

Yes, 180ms is roundtrip.
I understand that essentially I cannot do anything for WAN but
which test should I run exactly on my PC?

Another question: would a VPN connection beneficial for speed or is just useful to skip the 15 hops?

A VPN connection over the same network would make it slower, not faster.

nitro

 

[JackR]

You have three types of delay involved:

a)propagation delay - the speed of electrons \photons moving over cable or fiber

b)inherent router delay - the time each router takes to route(address to the next router) the packets you are sending

c)traffic loading delay - additional variable delay due to the amount of traffic being handled by all the routers in the network. Any one router could be heavily loaded and place your packet in a queue. This would slow things down. It accounts for the variation in the ping time you reported.

As Nitro stated, VPN use would slow things down. All you are essentially doing is adding an additional layer of software to the existing network.

The Internet uses a set of protocols specifically designed to allow sharing of a "backbone" data path. The backbone is made up of many different routers, typically interconnected in a mesh such that the loss of any router will cause the remaining routers to seek a new path to the desired destination. Protocols are used to select the "least-cost" path. Thus, by many users sharing the path, the cost is kept very low for individual users. If you really need a faster path, you could probably get it down to the 30-50 millisecond range. You would need a leased line between your location and your broker. I don't know where you are in Europe, but the cost would probably run in the $3,000 per month range for a fractional T1 line.

 

[nitro]

I just installed the 64-bit version of Microsoft Server 2003 for Extended Systems, downloaded the 32-bit version of Sandra, reran the memory bandwidth test and got 13.5 GB/s :eek: :eek: :eek: :eek: Consistently :eek: It now destroys all systems listed as "similar" to my setup

I also reran the mem/cache test and it came back with a Combined Index of 26263 MB/s and a Speed Factor of 6.6!!! :eek:

That gives me huge motivation to port over to 64-bits. My gawd, imagine if I ran the 64-bit version of Sandra :eek:

This machine SMOKES. But in order to really get at it, you have to run in 64-bits. Tyan once again was right - there is a critical part of the HyperTransport that is simply not in play in 32-bit mode.

nitro

 

[damir00]

if you care enough about speed to go through the trouble and expense of colocating, why are you running windows? the linux networking stack is hugely faster than that in any version of windows, switching would give you at least as much latency reduction as the move from residential to colo - and for a fraction of the price.

 

[nitro]

Quote from damir00:

if you care enough about speed to go through the trouble and expense of colocating, why are you running windows? the linux networking stack is hugely faster than that in any version of windows...

True, but ythe reason I do not run "Linux" is that I am at the mercy of my data providers and "Linux"/Unix is not supported. Otherwise I would have been running FreeBSD or some Linux distro long ago.

...switching would give you at least as much latency reduction as the move from residential to colo - and for a fraction of the price.

That is false. It is nowhere even in the same ballpark the latency I would save by going to a "linux" TCP/IP stack. For example, from my colocated server, the ping time to my broker is 1 milliseconds - I am not kidding. Further, I colocate for two reasons, 1) is latency, the other is 2) bandwidth.

nitro

 

[Runningbear]

Nitro,

Which remote desktop software do you use?

Runningbear

 

[nitro]

Quote from Runningbear:

Nitro,

Which remote desktop software do you use?

Runningbear

Runningbear,

I just use what comes with Windows. You can also download it free from the MSFT website.

nitro