[STDS-802-3-BWA] FW: [STDS-802-3-BWA] A Question on Latency

["Martin, David" <damartin@xxxxxxxxx>]

RESEND

 

From: Martin, David
Sent: Thursday, July 07, 2011 5:47 PM
To: 'Vipul Bhatt'; STDS-802-3-BWA@xxxxxxxxxxxxxxxxx
Subject: RE: [STDS-802-3-BWA] A Question on Latency

 

Hi Vipul,

 

Yes, FEC can be disabled on carefully crafted networks for financial applications (e.g., millisecond trading), New York – Chicago is a good example.

 

FEC latency obviously varies by scheme (higher coding gain generally increases latency) and rate. Some ballpark ranges: 10-100usec at OTU2/10Gig; 3-30usec at OTU3/40Gig.

 

…Dave

 

From: Vipul Bhatt [mailto:vjb@xxxxxxxxxxxxx]
Sent: Thursday, July 07, 2011 5:20 PM
To: STDS-802-3-BWA@xxxxxxxxxxxxxxxxx
Subject: [STDS-802-3-BWA] A Question on Latency

 

Dear colleagues,

 

In the last BWA conference call, if I recall correctly, someone stated that for applications like financial transactions, FEC is often turned off to reduce latency -- even for traffic between New York and Chicago.

 

Please help me understand that. Light propagation delay in fiber is ~5 microseconds per kilometer. In contrast, FEC  latency would be sub-microsecond. Even within a data center of 300 meter span, a round-trip through just one switch can add up to several microseconds. On the other hand, turning off FEC may increase BER and make TCP retransmissions more likely, thus adding to latency.

 

Or is it that in such links, the BER is so low that FEC ends up being all latency and no benefit?

 

Thank you.

 

Regards,

Vipul

 

Vipul Bhatt

Vipul_Bhatt@xxxxxxxx

(408) 461-8521