FTP Slow From Alpha to External Sites

[Robert Atkinson]

I've got a real quandry.

If I FTP from our Alpha's to an internal FTP server, I get an expected 5000 kbps.

If I then FTP to an external site, through our Firewall (Firewall-1) then I only get 30 kbps.

Ah, slow pipe I hear you cry. Well not really, cause of I sit on my PC and FTP to the same site, then I get 500 kbps.

I can't explain the difference. Why should one OS be faster or slower than another to the same site?

Rob.

[Ian Miller.]

I assume you have checked for the popular duplex mismatch problem?

Are you sending or receving a file to your alpha?

[Robert Atkinson]

Yeh, but the fact that we get 5000k rates internally rules that out.

I haven't tried pulling files from the Alpha yet - guess I could give it a try from home tonight.

Rob.

[Ian Miller.]

If you are pushing files to the alpha then have yuo tried increasing the extend quantity used for the files. If the files are big then the default file extend quantity is inadequate.

[Petr Spisek]

Hi, check route setting on your alphaserver and trace path to the external site from your PC and alphaserver.

Petr

[David Jones_21]

I'd guess the default send/receive buffer quotas on the VMS box are too small. The effect of the small quotas increases as the latency between the endpoints increases.

[Robert Atkinson]

The files are being pushed from the Alpha to a remote server (HP's as it happens).

Petr - why do you suggest checking the routing info. Is there a known problem with this?

As the problem is within the FTP Client, are there any window sizes/quotas that I can check?

Rob.

[Petr Spisek]

If I understand right, you have better throughput from your PC then AlphaServer in the same network segment (to an external site). Compare tcpip configuration between them (specially routes). Badly configurated route could take this slowing effect.
P.

[Robert Atkinson]

Checked and both machines go directly to our external router, so routes look fine.

Rob.

[John Yu_1]

maybe the working set limits are too low?

[Wim Van den Wyngaert]

Could you post a tcptrace/full ?
In it, the window size changes are shown.

Also post your sys$specific:[tcpip$etc]sysconfigtab.dat file.

Wim

[Robert Atkinson]

Wilm, with the Trace - do you litteraly want me to run this command, or are you looking for me to trace the packets going out to the remote FTP server?

ALPHA_ROB$$ typ sys$specific:[tcpip$etc]sysconfigtab.dat
#
# TCP/IP Services for OpenVMS V5.1
# sysconfig parameter file
#
nfs:
tcp_threads=8
udp_threads=8

vfs:
vnode_age=120

[Willem Grooters]

Petr's remark could wwll be the case - depending yout VMS/TCPIP version.
We had a similar problem, that we could drill down to two paths to a remote network (0.0.0.0) over different gateways (one the local NIC, the other the nearest router). We have the idea that starting the router would have introduces the second path but we never got proof. Deleting all routes and adding the right one solved the problem
(VMS 7.3-1, TCPIP 5.3, after clean install on new machine)

Willem

[Wim Van den Wyngaert]

Robert,

You have no special parameters set.

Before you do the slow ftp, open an extra session and start in it :
tcptrace/fu/pack=500/out=ip.txt xxx
where xxx is the ip address or name of the external site.
Then post ip.txt (with extention txt or I can't open it due to restrictions over here).

Wim

[Wim Van den Wyngaert]

Also try a traceroute to see if the same route is followed. This could be depending on segment size.

Wim

[Arch_Muthiah]

Bob,

The slow speed is posible due to CPU, the disk interface, and the overall system bandwidth in the external machine.

What type of machine/OS/Disk drives availbe in the external site. These are all consititute the reason for slow transfer rate.


Archunan

[Wim Van den Wyngaert]

Arcgu,

The system can not be in cause because ftp to an internal system has great performance.
So forget about cpu, working sets etc.

The remote system can not be blamed either because it is reacting fast when a PC is doing FTP.

IMO it must be something between the 2 nodes that is reacting differently (e.g. routers) and this only for a VMS node. So, the question is what is different in FTP-VMS compared with FTP-PC.

Robert : did you also check the tcp counters ?

Wim

[Arch_Muthiah]

Win

< The remote system can not be blamed
< either because it is reacting fast when a > PC is doing FTP.

Understand.. I am sorry.


Archunan

[Michael Yu_3]

Hi Rob,

Older versions of TCP/IP services for OpenVMS use window size of 4096 (UCX V4.2) while newer versions use window size of 61440 (TCP/IP V5.3, V5.4, etc.). FOr PC's, I think they might be using 128Kbytes window size.

Thanks and regards.

Michael

[Robert Atkinson]

The window size I saw in the trace was 4096. How can I change this in the FTP client, or do I need to do something in TCPIP$CONFIG?

Thanks for all your help BTW - I'll get the full trace done ASAP and pass back the results.

Rob.

[Robert Atkinson]

OK, stats from TCP/IP client are :-

TCP Source Port = 60990, TCP Destination Port = 20
TCP Sequence Number = 1774038205 = ^x69BDACBD
TCP Acknowledge Number = 551081025 = ^x20D8D441
Flags (URG=0,ACK=1,PSH=0,RST=0,SYN=0,FIN=0),
Window = 61440 = ^xF000
TCP Checksum = ^x931D, Urgent Pointer = 0 = ^x0000

Averages a window size of around 60,000.

ALPHA::RAA_RobertAtki0 09:58:13 TCPIP$FTP CPU=00:00:29.94 PF=60524 IO=174504 MEM=384
PUT (IMAGE) 860160 bytes 00:00:35.88 elapsed (23.41 KB/S)
Local: SYS$SYSROOT:[SYSEXE]SYSDUMP.DMP;3
Remote: test_tbs_sysdump3.dmp

Averages around 25KB/s


I also tried the Madgoat client, and results are similar.

Rob.

[Wim Van den Wyngaert]

You are doing a put, so the window size of the external node is the one that is important (you read and send, the external node receives in window size buffers and writes these to disk). NAs long as there is a lot of space in the window, the sender is not waiting for an ack.

Did you also check tracerout to see if the same rout is followed ?

A snapshot of the tcptrace could also be useful.

Wim

[Robert Atkinson]

Full trace attached

[Wim Van den Wyngaert]

Rob,

Could you post it as .txt ? Bank security ...

Wim

[Petr Spisek]

Rob, did you try traceroute from both comps to your external server? Did you compare number of hops and latency? When you checked routes.

I'm back with route config :-)

Petr