Evaluation case and Cuda streams

[Takis Panagopoulos]

Hello and well done for your excellent work! I would like to take some measurements of my own using PFAC in a evaluation case similar to that described at Accelerating Pattern Matching Using a Novel Parallel

Algorithm on GPU using snort rulles and DEFCON packets for my thesis.

Using Device 0: "GeForce GTX 760"

major = 3, minor = 0

----- time-driven version 

../test/pattern/snort_ruleset4

@@@@ profile PFAC_matchFromHost + texture ON 

The elapsed time is 450.410004 ms

The input size is 245235537 bytes

The throughput is 4.355774 Gbps

The number of matched is 71828400

I tried to lower the number of rules or the number of input to test but i get similar results

The throughput i achieve is relatively low compared to the one around 10Gbps reported in the paper. Is there anything obvious i am missing?

I use the cuda 5.5 toolkit (i had to change some parameters on the function involving the texture reference since it changed a bit)

Furthermore i read in some previous post that you intended to add cuda streams support. Is there a special reason they were not supported from the start (like the thread safety issue with the texture binding)?

If not i may attempt to add them myself and if the result is good i may contribute to the project.

Takis

[Cheng-Hung Lin]

Do you enhance the transmission of PCIe? You can try to use cudaMallocHost to allocate host memory instead of using malloc. 

In addition, the copy of match results from device to host is very time-consuming. If the number of pattern is less than 65535, please use "short int" for match result instead of "int".

Thanks for your response. We would appreciate if you contribute to the project.

--
這是 Google 網上論壇針對「pfacForum」群組發送的訂閱通知郵件。
如要取消訂閱這個群組並停止接收來自這個群組的郵件，請傳送電子郵件到 pfacForum+...@googlegroups.com。
如需更多選項，請前往：https://groups.google.com/d/optout。

--

Sincerely,
Cheng-Hung Lin, Associate Professor

Department of Electrical Engineering

National Taiwan Normal University, Taiwan
(O)886-2-77343432