Try TT Now

Registry Settings (Windows XP)

These are explanations of the additional NIC settings that are implemented per TT’s recommendations.

Registry Setting

TT Machine

Description

ArpCacheLife

Server Machines, Client Machines

This sets the timeout for unused entries in the OS ARP Cache. By default, this key does not exist but is set to two minutes. TT sets this to one (1) hour.

ArpCacheMinReferencedLife

Server Machines, Client Machines

This sets the timeout for all entries in the OS ARP Cache. By default, this key does not exist but is set to 10 minutes. TT sets this to two (2) hours.

Tcp1323Opts

Server Machines, Remote Hosts, WAN Routers

This allows a TCP receive window of larger than 64k to be advertised; it will only be seen during the SYN and will show up as a Windows Scaling multiplication factor. This is used when calculating the correct TCP Receive Window for a WAN connection.

EnableTCPA

Server Machines

This is Net DMA NetDMA and it allows for a Direct Memory Access (DMA) engine on the Peripheral Component Interconnect (PCI) bus. This does not work if RSS and TCPChimney are turned off, but Microsoft recommends disabling it.

EnableRSS

Server Machines

This toggles Receive Side Scaling. Receive Side Scaling only works on multi-processor boxes and allows the incoming data to be processed across multiple CPUs. TT turns this off as it causes out of order packets, duplicate packets, or dropped data. This really only applies to TCP.

EnableTCPChimney

Server Machines

This allows the OS to offload the TCP stack to the NIC. This is more efficient in that it allows the NIC CPU to process the TCP data and only pass up what is required. (This is only good in a streaming environment.) TT's data is small packets all with Push Bit set which means TT treats the TCP data more like a Packet-based Protocol - each packet is passed up to the applications. This is not an efficient use of offloading and does cause extra latency.

DisableTaskOffload

Server Machines

This toggles checksum offloading. In general, offloading the checksum calculation to the device driver is supposed to free up the OS from doing the work; therefore, making the process more efficient. TT has tested this, and it shows that leaving the OS to do the work is actually faster and more efficient.

IgnorePushBitOn Receives

Server Machines, Client Machines

This allows any incoming packets without Push Bit set to be treated as though they were sent that way. This makes the application more efficient by handling fragments; for example, say a 1500 byte packet turns up. Normally TCP would hold this until it receives what it assumes to be the end of the stream, or a packet with Push Bit set. This adds latency and can stall the application from handling the data because it is now waiting on the OS to pass it up. With this setting, TT's application can dissect the 1500 byte packet sent out - say 10 prices - but retain a small fragment while waiting for the next packet to arrive. This means that data is send all the time, making it a more consistent flow.

TcpAckFrequency

Server Machines

To make better use of the bandwidth the Nagle Algorithm is used. This means that it accumulates data until it receives an ACK from the receiving side; however, if delayed ACKs are turned on, it can create an even greater latency in sending and receiving data.

SynAttackProtect

Remote Hosts, WAN Routers

This closes half open TCP connections faster so a SYN attack cannot create a resource failure.

TcpTmedWaitDelay

Remote Hosts, WAN Routers

This decreases the time wait state for a closed TCP session. Normally the system takes three (3) minutes to recover the resource, but this can be set to 30 seconds.

TCPWindowSize

Remote Hosts, WAN Routers

This allows us to set the size of the TCP receive window to suit the bandwidth of the line. This is now normally set in the TTM config file.