MPE/iX Networking
Enhancements
Abstract
MPE/iX networking lab, has worked on a few enhancements and improvements to networking stack. Here, I would like to share some new enhancements and improvements done to the MPE/iX networking.
The main enhancements and improvements are on increasing some of the
TCP/UDP limits, implementation of traceroute utility, and resolving a TCP performance issue. The network limits increase include increasing total number of TCP connections and total number of UDP sockets. These enhancements will help our customers have increased connectivity to their HP 3000 machines. There have been many requests for the traceroute utility on 3000-L news group. We have developed traceroute utility which will help trace a route from one host to another also providing round-trip delay between source and various routers in between. This can be also be used to find out paths which are introducing large delays. TCP performance on MPE/IX is another issue and solution to the receive throughput problem is an information I would like to share along with other information mentioned above.
Increase in number of TCP connections
Currently, TCP on MPE/iX is allowing a maximum of 10240 TCP connections. Now we have made an enhancement to TCP module to go up to 20,000 connections. The version ID of TCP (NET_TCP_VERS in nmmaint,3) which supports this feature will start with 'B0605' (6.5 release).
Configuration through NMMGR
You need to configure the machine using NMMGR to allow it to go up to 20,000 connections. The field, "Maximum number of connections" can now handle up to 20,000 connections. Before this enhancement the range was from 0 to 10240.
To reach the required NMMGR screen do the following -
1. Go to [ Open Config ]
2. Go to [ NS ]
3. Go to [ Unguided ]
4. Go to [ Netxport ]
5. Go to [ Gprot ]
6. Go to [ Tcp ]
After the re-configuration don't forget to stop the network if it is already started and restart the network. Now TCP is ready to accept up to 20,000 TCP connections.
Increase in number of UDP Sockets
Currently, UDP on MPE/iX is allowing a maximum of 10240 sockets. Now we have made an enhancement to UDP module to get up to 10,000 sockets. The version ID of UDP (NET_UDP in nmmaint,3) which supports this feature will start with 'B0605' (6.5 release).
Configuration through NMMGR
You need to configure the machine using NMMGR to allow it to go up to 10,000 sockets. The field, "Maximum number of UDP sockets" can now handle up to 10,000 sockets. Before this enhancement the range was from 0 to 4096.
To reach the required NMMGR screen do the following -
1. Go to [ Open Config ]
2. Go to [ NS ]
3. Go to [ Unguided ]
4. Go to [ Netxport ]
5. Go to [ Gprot ]
6. Go to [ UDP ]
After the re-configuration don't forget to stop the network if it is already started and restart the network. Now UDP is ready to accept up to 10,000 UDP sockets.
Traceroute Utility on MPE/iX
There have been many requests for the traceroute utility on 3000-L news
group. The traceroute utility will help trace a route from one host to
another. The utility also provides round-trip delay between source and
various routers in between. This can be used to find out paths which
are introducing large delays.
Currently, the traceroute utility implemented on MPE/iX is undergoing
alpha testing. Attached below is an output of sample run of traceroute
utility (tracert) on MPE/iX -
:run tracert.group.acct
Enter Host Name[Press Return to exit]: pavan.india.hp.com
Resolving IP Address.....
Enter the number of hops[min=1,max=60]:
Defaulting to 25 hops
Traceroute to pavan.india.hp.com,hops=25
To TERMINATE Press CTRL-Y
#1 15.44.48.1 (atlagw2.atl.hp.com) 330ms 378ms 390ms
#2 15.44.48.1 (atlagw2.atl.hp.com) 401ms 448ms 459ms
#3 15.41.16.1 470ms 595ms 707ms
#4 15.24.240.2 (atlhgw2.cns.hp.com) 428ms 269ms 381ms
#5 15.88.32.1 (palhgw3.cns.hp.com) 585ms 711ms 826ms
#6 15.88.56.6 (palhgw6.cns.hp.com) 966ms 75ms 162ms
#7 15.111.31.2 (snghgw1.cns.hp.com) 453ms 769ms 294ms
#8 15.64.32.2 (snghgw2.cns.hp.com) 345ms 658ms 939ms
#9 15.10.32.1 (blrgw1.cns.hp.com) 794ms 817ms 902ms
#10 15.10.40.72 (pavan.india.hp.com) 796ms 828ms 793ms
END OF PROGRAM
The traceroute utility in its implementation makes use of TTL field of
IP header.
TCP Performance issue
It has been found using performance measurement utilities like XPPERF
and TTCP ( ported public domain utility ) that the receive throughput
on MPE/iX is considerably less compared to other platforms (eg. HP-UX).This problem has been traced to small default TCP receive window size on MPE/iX. This problem is now corrected.
Attached is the XPPERF output, taken between two machines named DRY and
HALIFAX with and without the fix.
DRY to HALIFAX (without fix)
----------------------------
Enter protocol to test (TCP, X25, UDP) :t
Enter socket number (def=32000) ..........
Enter transmission type (Unidirectional/Bidirectional): u
Enter mode (Master/Slave) : m
Enter remote node name : dry
Changes to run-time defaults? (Y/N).. :
BEGIN TCP master execution.
IPCCREATE ...
IPCCONTROL (socket timeout) ...
IPCDEST ...
IPCCONNECT ...
IPCSHUTDOWN (socket) ...
IPCCONTROL (connection timeout) ...
IPCRECV (connection) ...
Connection is now established.
Enter initial send size (def=1408) . : 100
Enter final send size (def=1408) . : 9100
Enter step size (def=1) ............ : 500
Enter # passes for each size (def=25): 2048
Hit RETURN when ready.
IPCSEND (for send info ...)
Start RECV loop [ 0]
Ending Buffer size Elapsed
step
time (bytes) Mbits/sec
KBytes/second time
---------- ----------- --------- ------------- -------
921005107 100 0.31 39.80 5
921005111 600 2.33 298.54 4
921005117 1100 3.42 437.85 5
921005124 1600 3.55 454.91 7
921005134 2100 3.27 417.95 10
921005145 2600 3.68 470.42 11
921005159 3100 3.44 440.69 14
921005175 3600 3.50 447.80 16
921005192 4100 3.75 480.00 17
921005212 4600 3.58 457.75 20
921005234 5100 3.60 461.37 22
921005257 5600 3.79 484.58 23
921005283 6100 3.79 485.62 25
921005311 6600 3.67 469.13 28
921005341 7100 3.81 487.26 29
921005373 7600 3.69 472.68 32
921005407 8100 3.70 474.14 34
921005444 8600 3.71 475.44 36
921005483 9100 3.63 464.39 39
DRY to HALIFAX (with fix)
-------------------------
Initial send size (def=1408) . : 100
Enter final send size (def=1408) . : 9100
Enter step size (def=1) ............ : 500
Enter # passes for each size (def=25): 2048
Hit RETURN when ready.
IPCSEND (for send info ...)
Start RECV loop [ 0]
Ending Buffer size Elapsed
step time (bytes) Mbits/sec KBytes/second time
---------- ----------- --------- ------------- -------
921082548 100 0.31 39.80 5
921082553 600 1.87 238.83 5
921082557 1100 4.28 547.31 4
921082563 1600 4.15 530.73 6
921082570 2100 4.66 597.07 7
921082577 2600 5.78 739.23 7
921082585 3100 6.03 771.22 8
921082594 3600 6.22 796.09 9
921082604 4100 6.37 816.00 10
921082616 4600 5.96 762.92 12
921082628 5100 6.61 845.85 12
921082641 5600 6.70 857.33 13
921082655 6100 6.77 867.17 14
921082671 6600 6.41 820.97 16
921082687 7100 6.90 883.17 16
921082704 7600 6.95 889.75 17
921082723 8100 6.63 848.47 19
921082742 8600 7.04 900.84 19
921082763 9100 7.07 905.56 20
Please note the increase in throughput. It is almost twice compared to
that before.
Summary
Over last one year MPE/iX Networking lab has worked on some enhancements which will benefit MPE/iX customers. They include increasing number of TCP connections, number of UDP sockets. This will help increased connectivity to MPE/iX machines. Other than these two enhancements related capacity increase, the traceroute utility was implemented on MPE/iX. Networking lab has worked on a TCP performance issue and resolved a problem related to TCP receive throughput.