packet loss
Brecht Vermeulen
brecht.vermeulen at UGent.be
Fri Jun 27 02:26:45 EDT 2003
Hi,
I'm not sure that tulip cards are much better. I think you would profit
more of a faster PCI bus (e.g. 64 bit - 66 MHz or faster).
I don't know if the CPU is the limit or what would be a possible upper
limit.
I can forward (in a router config) around 600kpps of 64 bytes and around
1.4 Gbps with 1400 byte packets, but I don't use click for packet
generation. We have a Spirentcom Smartbits for that.
What do you want to reach in fact ?
How much do you get with bigger packets ?
regards,
Brecht
Shilpi Agarwal wrote:
>
> Hi Brecht!!
>
> Well that made a lot of difference...but still i had to increase Tx Descriptor
> to get the high transmit rate ..and here are some of the results that i am
> getting for 64 byte packtes-
>
> Tx Max Loss Free rate
>
> 256 30k
> 512 65k
> 1000 100k
> 2000 250k
> 3000 300k
> 4000 400k
> 4096 450k
>
> And the maximum rate i can get is 522k with loss 3.6%
>
> Is there anythign else that can improve the performace ?? Can i get better
> performance with tulip cards??
>
> Thanks
> Shilpi
>
> Quoting Brecht Vermeulen <brecht.vermeulen at UGent.be>:
>
> >
> > Hi,
> >
> > can you try to add
> >
> > PollDevice(eth1) -> Discard;
> >
> > at the sending host also ?
> >
> > Don't know if this will make a difference though.
> >
> > regards,
> > Brecht
> >
> > > FastUDPSource(100000, 1000000, 1024,
> > > 00:02:b3:b6:c4:de, 192.168.0.1, 1234,
> > > 00:02:b3:b6:c5:31, 192.168.0.2, 1234)
> > > ->ToDevice(eth1);
> > >
> > > Receiving host:
> > >
> > > PollDevice(eth0,true)
> > > ->c0::Classifier(6/0002b3b6c4de,-)
> > >
> > > c0[0]
> > > ->Counter
> > > ->Discard;
> > >
> > > c0[1]
> > > ->Idle
> > > ->ToDevice(eth0);
> > >
> > > Quot
> > >
> > > ing Brecht Vermeulen <brecht.vermeulen at UGent.be>:
> > >
> > > >
> > > > Hi,
> > > >
> > > > hmmm, not sure that the card is the problem. I have no experience with
> > > > UDPsource. What happens if you use bigger packets ?
> > > >
> > > > What is the PCI bus to which the card is connected (64 bits - 66 MHz ?
> > > > or even higher ?) ?
> > > >
> > > > What is now the exact click configuration you use at both PCs ?
> > > >
> > > > I assume you're using the patched e1000 driver ?
> > > > Can you see something the /proc/PRO_LAN statistics ?
> > > >
> > > >
> > > > regards,
> > > > Brecht
> > > >
> > > > Shilpi Agarwal wrote:
> > > > >
> > > > > Hi Brecht!!
> > > > >
> > > > > Thank you for your email.I tried to check output of ifconfig after
> > each
> > > > test and
> > > > > i found that error is at the transmittor. Its not sending all the
> > packets.
> > > > >
> > > > > receiver is able to receive packtes without any loss. Then i increased
> > the
> > > > > number o Tx Descriptors at the transmittor and here are some of the
> > > > results
> > > > > for rate 100 thousand pps and packets size 64 bytes.
> > > > >
> > > > > Tx Descriptor Packets sent by UDp source Packtes sent by NIC
> > > > > 256 100,0000 996094
> > > > > 512 998098 998098
> > > > > 1024 999068 999068
> > > > > 2048 999540 999540
> > > > > 4096 999778 999778
> > > > >
> > > > > It clearly indicates that the problem is at NIC. I have Pentium 4 ,
> > 2.53
> > > > Ghz
> > > > > processor and network adaptor is Intel Pro/1000 T Desktop RJ45.
> > > > >
> > > > > I looked at message-
> > > > > http://www.pdos.lcs.mit.edu/click/ml/click/msg01900.html
> > > > >
> > > > > It looks like Tulip cards are faster tahn intel. Do you think i might
> > get
> > > > rid of
> > > > > this loss by using Tulip cards?? Or which card would be best ??
> > > > >
> > > > > Thanks
> > > > > Shilpi
> > > > >
> > > > > Quoting Brecht Vermeulen <brecht.vermeulen at UGent.be>:
> > > > >
> > > > > >
> > > > > > I'm not sure, but ifconfig says the number of packets (I don't know
> > if
> > > > > > those rejected are included or not) and some drivers, e.g. e1000
> > have
> > > > > > some statistics read of the cards in /proc/PRO_LAN (or something
> > like
> > > > > > that).
> > > > > >
> > > > > > don't know for sure. If you are connected to a switch, do you know
> > the
> > > > > > number then ?
> > > > > >
> > > > > > Maybe you can try a counter on the sending machine also if you
> > haven't
> > > > > > already.
> > > > > >
> > > > > > I don't know exactly what rejected packets mean. I have seen it here
> > on
> > > > > > non-click machines with cheap network cards (dlink e.g.)
> > > > > >
> > > > > > regards,
> > > > > > Brecht
> > > > > >
> > > > > > Shilpi Agarwal wrote:
> > > > > > >
> > > > > > > Hi,
> > > > > > >
> > > > > > > I ran dmesg on sending host and i saw the message eth1 rejected a
> > > > packet.
> > > > > > b
> > > > > > > Does this mean it rejected only one packet or lot of packets.
> > > > > > >
> > > > > > > If the machines are connected directly how can i measure the number
> > of
> > > > > > packets
> > > > > > > actually sent by device driver.
> > > > > > >
> > > > > > > Quoting Brecht Vermeulen <brecht.vermeulen at UGent.be>:
> > > > > > >
> > > > > > > >
> > > > > > > > Hi,
> > > > > > > >
> > > > > > > > and which packet length are you using ? (can you give the config
> > of
> > > > the
> > > > > > > > sending host)
> > > > > > > > Try using packets of e.g; 1400 bytes.
> > > > > > > >
> > > > > > > > Have you tried to connect both PCs without a switch (crossed
> > cable
> > > > ?)
> > > > > > > > Is it 10/100 ethernet ?
> > > > > > > >
> > > > > > > > Can you see statistics on the switch ? (to be sure that the
> > sending
> > > > host
> > > > > > > > is sending all packets)
> > > > > > > >
> > > > > > > > regards,
> > > > > > > > Brecht
> > > > > > > >
> > > > > > > > Shilpi Agarwal wrote:
> > > > > > > > >
> > > > > > > > > Hi,
> > > > > > > > >
> > > > > > > > > Thanks for your email, i have tried that
> > > > > > > > > I have tried with configuration like
> > > > > > > > >
> > > > > > > > > PollDevice(eth0,true)
> > > > > > > > > ->c0::Classifier(6/0002b3b6c4de,-)
> > > > > > > > >
> > > > > > > > > c0[0]
> > > > > > > > > ->Counter
> > > > > > > > > ->Discard;
> > > > > > > > >
> > > > > > > > > c0[1]
> > > > > > > > > ->Idle
> > > > > > > > > ->ToDevice(eth0);
> > > > > > > > >
> > > > > > > > > But i still see the same packet loss. At a rate till 15
> > thousand
> > > > pps
> > > > > > its
> > > > > > > > fine ,
> > > > > > > > > afterwards it starts lossing packets. At 100 thousand pps loss
> > is
> > > > .38%
> > > > > > and
> > > > > > > > it
> > > > > > > > > remains between .28 & .36% till 600 thousands pps, which is
> > > > maximum
> > > > > > sending
> > > > > > > > rate
> > > > > > > > > i could get.
> > > > > > > > >
> > > > > > > > > And other thing is Using one single interface and default value
> > of
> > > > > > > > transmit
> > > > > > > > > descriptor i could achieve maximum send rate of 385.356
> > thousand
> > > > pps.
> > > > > > > > > When i tried to use both the interfaces of a machine to send
> > > > packets
> > > > > > the
> > > > > > > > total
> > > > > > > > > maximum send rate i can get is 585 thousand pps.
> > > > > > > > >
> > > > > > > > > I think it should be double of 385 since each device works
> > > > > > independently.
> > > > > > > > >
> > > > > > > > > Thanks
> > > > > > > > > Shilpi
> > > > > > > > >
> > > > > > > > > Quoting Brecht Vermeulen <brecht.vermeulen at UGent.be>:
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Hi,
> > > > > > > > > >
> > > > > > > > > > and did you try Eddie's answer to your previous question ?
> > > > > > > > > > http://www.pdos.lcs.mit.edu/click/ml/click/msg02004.html
> > > > > > > > > >
> > > > > > > > > > (not sure if that is the problem though)
> > > > > > > > > >
> > > > > > > > > > regards,
> > > > > > > > > > Brecht
> > > > > > > > > >
> > > > > > > > > > Shilpi Agarwal wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hi,
> > > > > > > > > > >
> > > > > > > > > > > i did a small experiment to check the maximum packet
> > > > handling
> > > > > > rate.
> > > > > > > > > > >
> > > > > > > > > > > I have two machines running click.
> > > > > > > > > > >
> > > > > > > > > > > On one machine i have
> > > > > > > > > > > UDPsource and in other machine i just have
> > > > > > > > > > >
> > > > > > > > > > > PollDevice(eth0,true)
> > > > > > > > > > > ->Counter
> > > > > > > > > > > ->Discard;
> > > > > > > > > > >
> > > > > > > > > > > The two nodes are connected via switch.
> > > > > > > > > > >
> > > > > > > > > > > Now i am sending 100,0000 packets at a rate of 100 thousand
> > pps
> > > > but
> > > > > > at
> > > > > > > > the
> > > > > > > > > > sink
> > > > > > > > > > > i never receive all packets .Counter always show around
> > > > 996138.
> > > > > > > > > > > even if i reduce the sending rate to 20 thousand pps still
> > i
> > > > am
> > > > > > not
> > > > > > > > > > receiving
> > > > > > > > > > > all the packets . At 20 thousand pps i could receive only
> > > > 996181
> > > > > > > > packets.
> > > > > > > > > > >
> > > > > > > > > > > does anyone have an idea why i am loosing packets??
> > > > > > > > > > >
> > > > > > > > > > > Thanks
> > > > > > > > > > > Shilpi
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Shilpi Agarwal
> > > > > > > > > > > Graduate Student
> > > > > > > > > > > Dept of Electrical Engineering
> > > > > > > > > > > University of Wisconsin
> > > > > > > > > > > Madison-53706
> > > > > > > > > > > phone:608-233-3714
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > > Shilpi Agarwal
> > > > > > > > > Graduate Student
> > > > > > > > > Dept of Electrical Engineering
> > > > > > > > > University of Wisconsin
> > > > > > > > > Madison-53706
> > > > > > > > > phone:608-233-3714
> > > > > > > >
> > > > > > >
> > > > > > > Shilpi Agarwal
> > > > > > > Graduate Student
> > > > > > > Dept of Electrical Engineering
> > > > > > > University of Wisconsin
> > > > > > > Madison-53706
> > > > > > > phone:608-233-3714
> > > > > >
> > > > >
> > > > > Shilpi Agarwal
> > > > > Graduate Student
> > > > > Dept of Electrical Engineering
> > > > > University of Wisconsin
> > > > > Madison-53706
> > > > > phone:608-233-3714
> > > >
> > >
> > > Shilpi Agarwal
> > > Graduate Student
> > > Dept of Electrical Engineering
> > > University of Wisconsin
> > > Madison-53706
> > > phone:608-233-3714
> >
>
> Shilpi Agarwal
> Graduate Student
> Dept of Electrical Engineering
> University of Wisconsin
> Madison-53706
> phone:608-233-3714
More information about the click
mailing list