Thanks Roman, as you may have noticed these changes did get merged! Eddie
On 5/23/11 1:34 PM, Roman Chertov wrote: > Finally I had the time to test the changes the Cliff made to > To/FromUserDevice elements. I have tried the elements using on patched > 2.6.24.7 and Fedora14 patchless kernels. In both cases, I was able to > capture ~1Gbps in my userlevel application. The userlevel application > just reads the data from the ToUserDevice ring buffer and then discards > it. If the data cannot be read fast enough from ToUserDevice, then the > ring buffer would overflow and drop_count would indicate the number of > dropped packets. The stats shown below demonstrate that no drops > occurred and the entire stream was read by the user level app. However, > in order to achieve this performance I had to increase the BURST and > CAPACITY parameters. > > I recommend that the changes made by Cliff be merged with the mainline. > > > Receiver Stats > [rchertov@nm0:~/pkt_reader]$ more /click/ctr/rate > 82004 > [rchertov@nm0:~/pkt_reader]$ more /click/ctr/count > 833000 > [rchertov@nm0:~/pkt_reader]$ more /click/ToUserDevice\@3/drop_count > 0 > > > Receiver Config > pd :: FromDevice(eth21) > -> ctr :: AverageCounter > -> ToUserDevice(0, CAPACITY 512, ENCAP len32, BURST 10); > > Idle -> ToDevice(eth21); > > > Sender Config > > src1 :: RatedSource(\<010101010101010101010101010101010101010101010101>, > LENGTH > 1458, RATE 83000, LIMIT 833000) > -> UDPIPEncap(SRC 10.1.1.3, SPORT 6667, DST 10.1.1.7, DPORT 6667) > -> SetIPDSCP(46) > -> EtherEncap(0x0800, 00:15:17:7C:E5:15, 00:15:17:7C:E3:B1) > -> StoreUDPTimeSeqRecord(OFFSET 14, DELTA false) > -> ctr :: AverageCounter > -> ToDevice(eth1); > > > _______________________________________________ > click mailing list > [email protected] > https://amsterdam.lcs.mit.edu/mailman/listinfo/click _______________________________________________ click mailing list [email protected] https://amsterdam.lcs.mit.edu/mailman/listinfo/click
