Difference between revisions of "Hall-B to EJFat transfer speeds"

From epsciwiki
Jump to navigation Jump to search
 
(4 intermediate revisions by the same user not shown)
Line 1: Line 1:
  
 
== 2023-12-14 ==
 
== 2023-12-14 ==
* Beam down for a couple of hours so take advantage to run some bandwidth tests between Hall-B and EJFat in CC
+
Beam down for a couple of hours so take advantage to run some bandwidth tests between Hall-B and EJFat in CC.
  
 +
Some details are captured below but here is a summary:
 +
# iperf3 does not report rates as high as iperf(2)
 +
#* (I copied /usr/bin/iperf from efat-fs-daq to a temporary directory on clonfarm21sro to use it)
 +
# Can achieve 99Gbps between ejfat-2 and ejfat-fs-daq
 +
# Can achieve 75Gbps between clonfarm21sro and ejfat-fs-daq
 +
#* n.b. 75Gbps is a reasonably reliable average, but it fluctuates. The highest I saw was just over 80Gbps, but not sustained.
 +
# UDP is very limited (~1Mbps/thread). Possibly due to maximum allowed packet size on the system.
 +
# Routing from ejfat-fs-daq to clonfarm21sro is broken so only connections initiated from clonfarm21sro can be made at the moment.
 +
 +
 +
===TCP===
 
  clonfarm21:~>traceroute ejfat-fs-daq
 
  clonfarm21:~>traceroute ejfat-fs-daq
 
  traceroute to ejfat-fs-daq (129.57.177.7), 30 hops max, 60 byte packets
 
  traceroute to ejfat-fs-daq (129.57.177.7), 30 hops max, 60 byte packets
Line 8: Line 19:
 
   2  * * *
 
   2  * * *
 
   3  ejfat-fs-daq.jlab.org (129.57.177.7)  0.174 ms  0.157 ms *
 
   3  ejfat-fs-daq.jlab.org (129.57.177.7)  0.174 ms  0.157 ms *
 +
 +
n.b. this fails to find a route to ''clonfarm21sro'' from ejfat-fs-daq
 +
davidl@ejfat-fs:~$ traceroute clonfarm21sro
 +
traceroute to clonfarm21sro (129.57.69.11), 30 hops max, 60 byte packets
 +
  1  _gateway (129.57.177.129)  0.154 ms  0.156 ms  0.137 ms
 +
  2  hallb-ejfat-rtr-2023-172-31-28-29.jlab.org (172.31.28.29)  0.179 ms  0.211 ms  0.243 ms
 +
  3  * * *
 +
  4  * * *
 +
  5  * * *
  
 
Sending data via iperf3 to ejfat-fs-daq cannot seem to go faster than ~4Gbps. Same results for 1 thread and 8 threads:
 
Sending data via iperf3 to ejfat-fs-daq cannot seem to go faster than ~4Gbps. Same results for 1 thread and 8 threads:
Line 79: Line 99:
 
  [  5]  2.00-4.00  sec  4.66 GBytes  20.0 Gbits/sec    0    430 KBytes       
 
  [  5]  2.00-4.00  sec  4.66 GBytes  20.0 Gbits/sec    0    430 KBytes       
 
  [  5]  4.00-6.00  sec  4.70 GBytes  20.2 Gbits/sec    0    430 KBytes
 
  [  5]  4.00-6.00  sec  4.70 GBytes  20.2 Gbits/sec    0    430 KBytes
 +
 +
I am able to get faster transfer speeds using iperf (=iperf2) than I can with iperf3. Running this command on ejfat-2 gives full bandwidth (up to 99Gbps). Running on clonfarm21 though only gets up to around 50Gbps. Note that I edited out individual thread lines from the following for brevity.
 +
 +
clonfarm21:~/tmp>./iperf -c ejfat-fs-daq -t 3600 -i4 -P8 -w 500k
 +
------------------------------------------------------------
 +
Client connecting to ejfat-fs-daq, TCP port 5001
 +
TCP window size:  416 KByte (WARNING: requested  488 KByte)
 +
------------------------------------------------------------
 +
[ ID] Interval      Transfer    Bandwidth
 +
[SUM]  0.0- 4.0 sec  18.4 GBytes  39.6 Gbits/sec
 +
[SUM]  4.0- 8.0 sec  20.1 GBytes  43.1 Gbits/sec
 +
[SUM]  8.0-12.0 sec  22.3 GBytes  47.9 Gbits/sec
 +
[SUM] 12.0-16.0 sec  25.3 GBytes  54.4 Gbits/sec
 +
[SUM] 16.0-20.0 sec  25.1 GBytes  53.9 Gbits/sec
 +
[SUM] 20.0-24.0 sec  25.0 GBytes  53.8 Gbits/sec
 +
[SUM] 24.0-28.0 sec  25.3 GBytes  54.4 Gbits/sec
 +
[SUM] 28.0-32.0 sec  25.1 GBytes  54.0 Gbits/sec
 +
[SUM]  0.0-32.8 sec  192 GBytes  50.2 Gbits/sec
 +
 +
Increasing the number of threads and playing with the window and buffer sizes gets up to around 75Gbps (avg)
 +
clonfarm21:~/tmp>./iperf -c ejfat-fs-daq -t 3600 -i4 -P24 -w 416K -l512K
 +
------------------------------------------------------------
 +
Client connecting to ejfat-fs-daq, TCP port 5001
 +
TCP window size:  416 KByte
 +
------------------------------------------------------------
 +
[ ID] Interval      Transfer    Bandwidth
 +
[SUM]  0.0- 4.0 sec  32.9 GBytes  70.7 Gbits/sec
 +
[SUM]  4.0- 8.0 sec  36.5 GBytes  78.4 Gbits/sec
 +
[SUM]  8.0-12.0 sec  35.5 GBytes  76.2 Gbits/sec
 +
[SUM] 12.0-16.0 sec  36.8 GBytes  79.1 Gbits/sec
 +
[SUM] 16.0-20.0 sec  35.1 GBytes  75.4 Gbits/sec
 +
[SUM] 20.0-24.0 sec  36.2 GBytes  77.7 Gbits/sec
 +
[SUM] 24.0-28.0 sec  33.9 GBytes  72.8 Gbits/sec
 +
[SUM]  0.0-29.4 sec  259 GBytes  75.7 Gbits/sec
 +
 +
=== UDP ===
 +
Sending via UDP is severely limited. Buffer lengths must be kept below the maximum allowed (<64KB) or they will not be sent. It also seems to be limited to ~1Mbps/thread (excluding the first set of packets which always register ~1.25Mbps). Running with 64 threads leads to a steady state of only ~65Mbps aggregate.
 +
 +
./iperf -c ejfat-fs-daq -t 3600 -i4 -l62K -u -P64
 +
------------------------------------------------------------
 +
Client connecting to ejfat-fs-daq, UDP port 5001
 +
Sending 63488 byte datagrams, IPG target: 484375.00 us (kalman adjust)
 +
UDP buffer size:  208 KByte (default)
 +
------------------------------------------------------------
 +
[SUM]  0.0- 4.0 sec  38.8 MBytes  81.3 Mbits/sec
 +
[SUM]  4.0- 8.0 sec  31.0 MBytes  65.0 Mbits/sec

Latest revision as of 13:21, 14 December 2023

2023-12-14

Beam down for a couple of hours so take advantage to run some bandwidth tests between Hall-B and EJFat in CC.

Some details are captured below but here is a summary:

  1. iperf3 does not report rates as high as iperf(2)
    • (I copied /usr/bin/iperf from efat-fs-daq to a temporary directory on clonfarm21sro to use it)
  2. Can achieve 99Gbps between ejfat-2 and ejfat-fs-daq
  3. Can achieve 75Gbps between clonfarm21sro and ejfat-fs-daq
    • n.b. 75Gbps is a reasonably reliable average, but it fluctuates. The highest I saw was just over 80Gbps, but not sustained.
  4. UDP is very limited (~1Mbps/thread). Possibly due to maximum allowed packet size on the system.
  5. Routing from ejfat-fs-daq to clonfarm21sro is broken so only connections initiated from clonfarm21sro can be made at the moment.


TCP

clonfarm21:~>traceroute ejfat-fs-daq
traceroute to ejfat-fs-daq (129.57.177.7), 30 hops max, 60 byte packets
 1  gw-69.jlab.org (129.57.69.1)  0.169 ms  0.202 ms  0.226 ms
 2  * * *
 3  ejfat-fs-daq.jlab.org (129.57.177.7)  0.174 ms  0.157 ms *

n.b. this fails to find a route to clonfarm21sro from ejfat-fs-daq

davidl@ejfat-fs:~$ traceroute clonfarm21sro
traceroute to clonfarm21sro (129.57.69.11), 30 hops max, 60 byte packets
 1  _gateway (129.57.177.129)  0.154 ms  0.156 ms  0.137 ms
 2  hallb-ejfat-rtr-2023-172-31-28-29.jlab.org (172.31.28.29)  0.179 ms  0.211 ms  0.243 ms
 3  * * *
 4  * * *
 5  * * *

Sending data via iperf3 to ejfat-fs-daq cannot seem to go faster than ~4Gbps. Same results for 1 thread and 8 threads:

Connecting to host ejfat-fs-daq, port 5201
[  5] local 129.57.69.11 port 41206 connected to 129.57.177.7 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-2.00   sec   924 MBytes  3.88 Gbits/sec    0    683 KBytes       
[  5]   2.00-4.00   sec   895 MBytes  3.75 Gbits/sec    0    683 KBytes       
[  5]   4.00-6.00   sec   889 MBytes  3.73 Gbits/sec    7    543 KBytes       
[  5]   6.00-7.05   sec   464 MBytes  3.70 Gbits/sec    0    543 KBytes       

clonfarm21:~>iperf3 -c ejfat-fs-daq -t3600 -i2 -P8
Connecting to host ejfat-fs-daq, port 5201
[  5] local 129.57.69.11 port 52242 connected to 129.57.177.7 port 5201
[  7] local 129.57.69.11 port 52252 connected to 129.57.177.7 port 5201
[  9] local 129.57.69.11 port 52258 connected to 129.57.177.7 port 5201
[ 11] local 129.57.69.11 port 52268 connected to 129.57.177.7 port 5201
[ 13] local 129.57.69.11 port 52276 connected to 129.57.177.7 port 5201
[ 15] local 129.57.69.11 port 52286 connected to 129.57.177.7 port 5201
[ 17] local 129.57.69.11 port 52300 connected to 129.57.177.7 port 5201
[ 19] local 129.57.69.11 port 52308 connected to 129.57.177.7 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-2.00   sec   121 MBytes   507 Mbits/sec    0    375 KBytes       
[  7]   0.00-2.00   sec   122 MBytes   514 Mbits/sec    0    546 KBytes       
[  9]   0.00-2.00   sec   120 MBytes   503 Mbits/sec    0    335 KBytes       
[ 11]   0.00-2.00   sec   120 MBytes   504 Mbits/sec    0    324 KBytes       
[ 13]   0.00-2.00   sec   120 MBytes   502 Mbits/sec    0    294 KBytes       
[ 15]   0.00-2.00   sec   120 MBytes   503 Mbits/sec    0    313 KBytes       
[ 17]   0.00-2.00   sec   120 MBytes   502 Mbits/sec    0    308 KBytes       
[ 19]   0.00-2.00   sec   120 MBytes   502 Mbits/sec    0    314 KBytes       
[SUM]   0.00-2.00   sec   963 MBytes  4.04 Gbits/sec    0             
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5]   2.00-4.00   sec   107 MBytes   448 Mbits/sec    0    375 KBytes       
[  7]   2.00-4.00   sec   108 MBytes   451 Mbits/sec    0    546 KBytes       
[  9]   2.00-4.00   sec   108 MBytes   451 Mbits/sec    0    335 KBytes       
[ 11]   2.00-4.00   sec   108 MBytes   451 Mbits/sec    0    324 KBytes       
[ 13]   2.00-4.00   sec   108 MBytes   451 Mbits/sec    0    294 KBytes       
[ 15]   2.00-4.00   sec   107 MBytes   450 Mbits/sec    0    313 KBytes       
[ 17]   2.00-4.00   sec   108 MBytes   453 Mbits/sec    0    308 KBytes       
[ 19]   2.00-4.00   sec   107 MBytes   450 Mbits/sec    0    314 KBytes       
[SUM]   2.00-4.00   sec   860 MBytes  3.61 Gbits/sec    0             
- - - - - - - - - - - - - - - - - - - - - - - - -
[  5]   4.00-6.00   sec   110 MBytes   461 Mbits/sec    0    375 KBytes       
[  7]   4.00-6.00   sec   110 MBytes   461 Mbits/sec    0    546 KBytes       
[  9]   4.00-6.00   sec   109 MBytes   459 Mbits/sec    0    335 KBytes       
[ 11]   4.00-6.00   sec   109 MBytes   459 Mbits/sec    0    324 KBytes       
[ 13]   4.00-6.00   sec   110 MBytes   460 Mbits/sec    0    294 KBytes       
[ 15]   4.00-6.00   sec   110 MBytes   461 Mbits/sec    0    313 KBytes       
[ 17]   4.00-6.00   sec   109 MBytes   457 Mbits/sec    0    308 KBytes       
[ 19]   4.00-6.00   sec   109 MBytes   458 Mbits/sec    0    314 KBytes       
[SUM]   4.00-6.00   sec   876 MBytes  3.68 Gbits/sec    0


Setting the window size higher does get faster rate. Note that if I set it higher than 416K it will return an error that the window size is not set correctly.

clonfarm21:~>iperf3 -c ejfat-fs-daq -t3600 -i2 -w 416K
Connecting to host ejfat-fs-daq, port 5201
[  5] local 129.57.69.11 port 47358 connected to 129.57.177.7 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-2.00   sec  4.15 GBytes  17.8 Gbits/sec    0    436 KBytes       
[  5]   2.00-4.00   sec  4.17 GBytes  17.9 Gbits/sec    1    436 KBytes       
[  5]   4.00-6.00   sec  4.16 GBytes  17.9 Gbits/sec    0    436 KBytes       

The default buffer length is 128k and the maximum size it allows you to set to is 1024k. This with the maximum window width gives the fastest rate of ~20Gbps. Note that THIS IS THAT SAME WHEN RUNNING THE CLIENT ON EJFAT-2!.

clonfarm21:~>iperf3 -c ejfat-fs-daq -t3600 -i2 -w 416K -l 1024K
Connecting to host ejfat-fs-daq, port 5201
[  5] local 129.57.69.11 port 36066 connected to 129.57.177.7 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-2.00   sec  4.68 GBytes  20.1 Gbits/sec    0    430 KBytes       
[  5]   2.00-4.00   sec  4.66 GBytes  20.0 Gbits/sec    0    430 KBytes       
[  5]   4.00-6.00   sec  4.70 GBytes  20.2 Gbits/sec    0    430 KBytes

I am able to get faster transfer speeds using iperf (=iperf2) than I can with iperf3. Running this command on ejfat-2 gives full bandwidth (up to 99Gbps). Running on clonfarm21 though only gets up to around 50Gbps. Note that I edited out individual thread lines from the following for brevity.

clonfarm21:~/tmp>./iperf -c ejfat-fs-daq -t 3600 -i4 -P8 -w 500k
------------------------------------------------------------
Client connecting to ejfat-fs-daq, TCP port 5001
TCP window size:  416 KByte (WARNING: requested  488 KByte)
------------------------------------------------------------
[ ID] Interval       Transfer     Bandwidth
[SUM]  0.0- 4.0 sec  18.4 GBytes  39.6 Gbits/sec
[SUM]  4.0- 8.0 sec  20.1 GBytes  43.1 Gbits/sec
[SUM]  8.0-12.0 sec  22.3 GBytes  47.9 Gbits/sec
[SUM] 12.0-16.0 sec  25.3 GBytes  54.4 Gbits/sec
[SUM] 16.0-20.0 sec  25.1 GBytes  53.9 Gbits/sec
[SUM] 20.0-24.0 sec  25.0 GBytes  53.8 Gbits/sec
[SUM] 24.0-28.0 sec  25.3 GBytes  54.4 Gbits/sec
[SUM] 28.0-32.0 sec  25.1 GBytes  54.0 Gbits/sec
[SUM]  0.0-32.8 sec   192 GBytes  50.2 Gbits/sec

Increasing the number of threads and playing with the window and buffer sizes gets up to around 75Gbps (avg)

clonfarm21:~/tmp>./iperf -c ejfat-fs-daq -t 3600 -i4 -P24 -w 416K -l512K
------------------------------------------------------------
Client connecting to ejfat-fs-daq, TCP port 5001
TCP window size:  416 KByte
------------------------------------------------------------
[ ID] Interval       Transfer     Bandwidth
[SUM]  0.0- 4.0 sec  32.9 GBytes  70.7 Gbits/sec
[SUM]  4.0- 8.0 sec  36.5 GBytes  78.4 Gbits/sec
[SUM]  8.0-12.0 sec  35.5 GBytes  76.2 Gbits/sec
[SUM] 12.0-16.0 sec  36.8 GBytes  79.1 Gbits/sec
[SUM] 16.0-20.0 sec  35.1 GBytes  75.4 Gbits/sec
[SUM] 20.0-24.0 sec  36.2 GBytes  77.7 Gbits/sec
[SUM] 24.0-28.0 sec  33.9 GBytes  72.8 Gbits/sec
[SUM]  0.0-29.4 sec   259 GBytes  75.7 Gbits/sec

UDP

Sending via UDP is severely limited. Buffer lengths must be kept below the maximum allowed (<64KB) or they will not be sent. It also seems to be limited to ~1Mbps/thread (excluding the first set of packets which always register ~1.25Mbps). Running with 64 threads leads to a steady state of only ~65Mbps aggregate.

./iperf -c ejfat-fs-daq -t 3600 -i4 -l62K -u -P64
------------------------------------------------------------
Client connecting to ejfat-fs-daq, UDP port 5001
Sending 63488 byte datagrams, IPG target: 484375.00 us (kalman adjust)
UDP buffer size:  208 KByte (default)
------------------------------------------------------------
[SUM]  0.0- 4.0 sec  38.8 MBytes  81.3 Mbits/sec
[SUM]  4.0- 8.0 sec  31.0 MBytes  65.0 Mbits/sec