| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1532.1 | | SMURF::RAJUL | | Fri Dec 16 1994 16:35 | 12 |
| >Are there a performance information of DEFEA, DEFPA used in 2100 server and
>OSF/1(TCP/IP)?
Actually, DEFPA is better than DEFEA for all protocols on transmits.
The EISA option can only transmit at about 75 Mbits/sec. The PCI
will do a full 100 Mbits/sec. Both boards can receive at full speeds.
This is true on other Alpha PCI-EISA systems. And I believe is true
for other operating systems (OpenVMS, for example).
Rajul
|
| 1532.2 | White paper available? | OSAV21::IZUTANI | Kenji Izutani, NIS Consul., DEC-Japan | Tue Dec 20 1994 03:09 | 6 |
| Thanks Rajul,
By the way, is there a "white paper" like Digital Technical Journal in which
the turbochannel performance was reported?
Kenji Izutani
|
| 1532.3 | | NETCAD::STEFANI | Look Ma, I'm drinking my oatmeal! | Tue Dec 20 1994 09:31 | 6 |
| >>By the way, is there a "white paper" like Digital Technical Journal in which
>>the turbochannel performance was reported?
Not today, but there are tentative plans to have one.
/l
|
| 1532.4 | | SMURF::RAJUL | | Thu Jan 05 1995 10:35 | 8 |
| >>By the way, is there a "white paper" like Digital Technical Journal in which
>>the turbochannel performance was reported?
Vol. 5 No. 1 Winter 1993 discusses DEFTA performance. The title of
the paper is:
"High-Performance TCP/IP and UDP/IP Networking in DEC OSF/1 for
Alpha AXP"
|
| 1532.5 | Performance data | SMURF::RAJUL | | Thu Jan 05 1995 10:45 | 37 |
| Here's performance data for DEFEA and DEFPA. Thanks to
Rick Formalaire in the Unix Performance Group. Note that
DEFPA performs better than DEFEA and V3.2 is better than
V3.0:
+-------+---------------+---------------+---------------+
| | Pair of 2100 | Pair of 2100 | Pair of 2100 |
| | with DEFEA | with DEFPA | with DEFEA |
| | FDDI Adapters,| FDDI Adapters,| FDDI Adapters,|
| | Full Duplex, | Full Duplex, | Full Duplex, |
| | OSF V3.2 (BL4)| OSF V3.2 (BL4)| OSF V3.0(BL12)|
| +-------+-------+-------+-------+-------+-------+
|Message| TCP | TCP | TCP | TCP | TCP | TCP |
| Size | Bandw | TX | Bandw | TX | Bandw | TX |
| | MB/s | CPU % | MB/s | CPU % | MB/s | CPU % |
+-------+-------+-------+-------+-------+-------+-------+
| 64 | 2.765 | 89.41 | 2.735 | 79.76 | 1.314 | 76.75 |
| 108 | 3.658 | 83.83 | 3.967 | 86.62 | 2.137 | 77.09 |
| 112 | 4.361 | 96.23 | 4.117 | 87.33 | 2.020 | 79.40 |
| 224 | 6.396 | 93.41 | 6.508 | 86.60 | 3.493 | 80.37 |
| 512 | 8.146 | 92.46 | 9.263 | 87.26 | 5.055 | 85.68 |
| 1024 | 8.847 | 74.37 | 9.005 | 66.59 | 6.352 | 86.76 |
| 1460 | 8.740 | 71.17 | 8.894 | 63.81 | 6.420 | 87.18 |
| 2048 | 8.857 | 66.93 | 9.021 | 57.22 | 7.094 | 90.85 |
| 3072 | 8.953 | 58.93 | 9.134 | 49.19 | 8.371 | 90.54 |
| 4096 | 8.956 | 51.47 | 9.089 | 47.99 | 8.621 | 91.33 |
| 6144 | 8.942 | 51.53 | 9.065 | 47.33 | 8.771 | 85.41 |
| 8192 | 8.953 | 49.74 | 9.004 | 46.21 | 8.778 | 82.06 |
| 10240 | 8.955 | 52.06 | 9.145 | 46.98 | 8.700 | 89.24 |
| 12288 | 8.939 | 50.27 | 9.046 | 45.93 | 8.801 | 86.56 |
| 14336 | 8.957 | 47.36 | 9.231 | 45.72 | 8.665 | 85.60 |
| 16384 | 8.884 | 45.70 | 9.420 | 46.93 | 8.577 | 82.18 |
| 17408 | 8.959 | 48.36 | 9.606 | 48.26 | 8.484 | 82.75 |
| 18432 | 8.947 | 49.47 | 9.409 | 48.01 | 8.365 | 82.16 |
| 32768 | 8.931 | 47.77 | 9.544 | 46.02 | 8.447 | 81.94 |
| 65536 | 8.959 | 45.65 | 9.541 | 45.89 | 8.473 | 81.97 |
+-------+-------+-------+-------+-------+-------+-------+
|
| 1532.6 | | OSAV21::IZUTANI | Kenji Izutani, NIS Consul., DEC-Japan | Thu Jan 05 1995 18:39 | 7 |
| Thanks again, Rajul,
.5 shows that even DEFPA w/3.2 do around 70Mbps.
Earlier reply implies this should be full 100Mbps. I'm confusing...
Kenji
|
| 1532.7 | Possible explanation | SMURF::RAJUL | | Tue Jan 10 1995 09:30 | 52 |
| >.5 shows that even DEFPA w/3.2 do around 70Mbps.
The best number was 76.8 Mbps. I suspect the TCP window
here was not optimum. Will have to check with our test
engineer and see if that is the case. I have seen close
to 95 Mbps with TCP/IP on our AlphaStations 200 and 400.
I used 128 Kb as the window-size. (See attached results
of some quick testing I had done a few months ago).
>Earlier reply implies this should be full 100Mbps. I'm confusing...
Apologies about the confusion. What I meant to say was that
under the best circumstances, the DEFEA is limited to 75 Mbps
while the DEFPA can transmit at speed. The "best" numbers
are seen with UDP/IP using optimal transfer sizes.
TCP/IP DEFPA/Avanti DEFPA/Mustang
Packet Size Xmt Thruput Xmt CPU Rcv Thruput Rcv CPU
----------- ----------- ------- ----------- -------
128 51.8 93.4 50.9 28.4
512 67.8 69.5 53.9 31.3
1024 80.9 64.9 80.5 48.7
1460 88.8 75.5 88.4 53.1
2048 87.9 63.7 83.4 50.0
4096 90.8 54.9 86.0 52.3
8192 92.9 55.2 92.6 57.7
16384 93.9 55.4 93.6 57.0
32768 94.3 57.3 94.3 57.8
65536 94.8 58.3 94.8 57.6
TCP/IP DEFPA/Mustang DEFPA/Avanti
Packet Size Xmt Thruput Xmt CPU Rcv Thruput Rcv CPU
----------- ----------- ------- ----------- -------
128 26.3 64.4 18.8 7.2
512 56.8 74.1 56.6 14.5
1024 71.3 79.3 70.3 30.0
1460 73.3 80.9 73.3 28.0
2048 84.8 85.9 76.8 31.5
4096 91.3 69.9 91.1 36.3
8192 93.0 67.5 93.0 37.7
16384 93.8 65.9 93.7 39.2
32768 94.3 66.5 94.3 37.9
65536 94.4 69.2 94.4 38.5
Rajul
|
| 1532.8 | 50% cpu utilization? | HURON::LINNELL | | Thu Feb 16 1995 15:10 | 7 |
| The HS1xx says it can transmit at 12MB/sec but this says the top speed for the
DEFEA which is used in the HS1xx is 8.9MB/sec. Moreover, it shows that 50% of
the Alpha CPU is burned transmitting at this speed. That's nearly 100MHz! Can
that be right? How can data be moved from disk to memory thru CPU to FDDI at
12MB/sec if mere memory to FDDI is 50% of CPU at 8.9 MB/sec?
/Andrew
|