Re: [ntdev] Sharing memory between kernel and user mode program

NTDEV
1

And your UM process can generate 20 million PPS of useful data? And you think that your NICs can transmit that fast? And what device will consume this flow?

I suspect that what you have is an architecture where the application is split between KM and UM and these packets are actually function calls between the two pieces. The obvious answer is to move everything to UM - you might be surprised how well it does even if you need a KM lower edge to talk to some hardware

One think to remember when porting from*NIX to Windows or the reverse is that often it is less work to implement your application in a paradigm that works with the platform than to port the source without changes and try to force a square peg into a round hole.

Sent from Surface Pro

From: xxxxx@gmail.com
Sent: ‎Wednesday‎, ‎April‎ ‎29‎, ‎2015 ‎2‎:‎42‎ ‎PM
To: Windows System Software Devs Interest List

For the performance I need to exchange about 20 millions packets per second, and every packet is 1500 bytes; I’ve seen that with DIRECT_IO ioctl I can archieve from 250000 to 400000 pps if I’m lucky.
I know that is a terrible way to work, I’ve read some articles and every article points out that; but unfortunately I cannot change the underlying architecture…

NTDEV is sponsored by OSR

Visit the list at: http://www.osronline.com/showlists.cfm?list=ntdev

OSR is HIRING!! See http://www.osr.com/careers

For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars

To unsubscribe, visit the List Server section of OSR Online at http://www.osronline.com/page.cfm?name=ListServer

2

What about future chipsets with presumably equally increases CPU and memory bandwidth to your faster NIC that presumably obviate the need to hack and slash?

There already exist 40 & 100 Gb/s NICs and Windows is quite good at using the bandwidth as long as there are many independent connections. UDP performance is also quite good, but a single TCP connection suffers from the overhead of maintaining the connection state

Sent from Surface Pro

From: xxxxx@gmail.com
Sent: ‎Thursday‎, ‎April‎ ‎30‎, ‎2015 ‎3‎:‎37‎ ‎PM
To: Windows System Software Devs Interest List

I know, it’s line rate (in fact 1.3 it’s more than line rate); but with multiple NICs I can go easily over; and can came out the need to inspect the packets inside the machine before letting them out.
Not mentioning that the program must be ready for NICs that will go beyond 10 GbE.

NTDEV is sponsored by OSR

Visit the list at: http://www.osronline.com/showlists.cfm?list=ntdev

OSR is HIRING!! See http://www.osr.com/careers

For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars

To unsubscribe, visit the List Server section of OSR Online at http://www.osronline.com/page.cfm?name=ListServer