File copying slow in network

[trilok_pvp]

Hi All,

I have 3 HP UX servers.

I am facing a backup problem in HP openview backup. When i tried to copy the file from a Node1 to Node2, it takes 1 min to transfer the 12MB file. But the same file when i try to copy it takes 10 to 15 mins form Node 2 to Node1.

[Pete Randall]

Run landiag against the NIC in Node2 (might as well check Node1 while you're at it) and look for collisions or other errors which might indicate a duplex mismatch.


Pete

[Michal Kapalka (mikap)]

hi,

check if you card have the same speed/duplex.

mikap

[trilok_pvp]

Hi,

I am in remote, How i can check the server in same duplex, any command to check the same.

[Michal Kapalka (mikap)]

hi,

use lanadmin and check the speed.

mikap

[Pete Randall]

See above: landiag.

Then you'll have to check the switch and how you check that depends on the switch.


Pete

[trilok_pvp]

Hi i check all are configuration is same.

Earlier its works fine. last few days only it shows the problem

[Michal Kapalka (mikap)]

hi,

there could be some change on Network layout, ask the Net Admins for some changes.

mikap

[trilok_pvp]

But if between the same servers, one Node1 server to other server, it is fast. From Node2 to Node1 it is show.

[Rita C Workman]

Yes, but remember that going out to the network is one thing. But connecting TO another servers' network port is another thing. Check your lancard port speeds. Confirm that they are correct for the card and that they are the same on the switch's port. Then check with your Network Admin that the switch's/router's port is set the same.
If these are not harmonious you will get flakey connections (if any connection at all).

Also, although not knowing your network, it is possible that the traffic in one direction is taking a different route then the traffic in the other direction. This too can affect your speed since you may be traversing different routers and/or making more hops.
Try doing a trace route from both sides and see what way the traffic is going.

Regards,
Rita

[rick jones]

Starting from the "top" as it were I would:

1) run something like netperf TCP_STREAM (www.netperf.org) from Node1 to Node2 and then from Node2 to Node1 and see if they can achieve the same level of performance each way.

2) check netstat -s -p tcp statistics when transferring in each direction - take them on the sending side. snapshot from before the copy/netperf then snapshot from after and run them through beforeafter:

netstat -s -p tcp > before
netperf or filecopy
netstat -s -p tcp > after
beforeafter before after > delta

check delta for retransmissions. beforeafter comes from ftp://ftp.cup.hp.com/dist/networking/tools/

If there are retransmissions, do the aforementioned lanadmin checks - use lanadmin to check the link-level statistics for the NIC(s) in use for the transfers - look for packet drops, CRC errors, frame errors etc.

If there were retransmissions, but the lanadmin *statistics* (statistics, not configuration being the same...) are "clean" then check the statistics on the switch(es) between the nodes.

If there are routers, also check the stats on the routers.

If netperf performs well each way, but the backup does not, check the filesystem performance on each node (read and write) via your favorite tool(s).

[trilok_pvp]

Hi All,

I had checked the LAN settings for the all the unix servers, all rae same. I had done some testing, when i copy a file from NodeA to any of the Unix servers and Windows servers, it is copying fast. When i try the same from any of the other server to this NodeA, its getting delay. Its Transfers FAST, Only delay in getting the file.

[rick jones]

You keep saying "settings" and we keep asking about "statistics" which are different things. Just to be certain there is no translation confusion, could you post the statistics from lanadmin and netstat as described prior?

[trilok_pvp]

Hi rick,

I had taken a statics of the Lan and Lanadmin config. I had not found any obnormalities in the file.

can u pls suggest on this.

Because the hp openview backup is installed in the server NODEA. When i take backup of this server its works fine. The other Server it takes lot of time. 1MB file in 1Min time.

Pls find the attachemnet on details on lanadmin and statics

[trilok_pvp]

Hi

I check with my system admin, he informed that no issues in the port nor the speed (switch reports 100/full on the connection.

It may be with FTP service, is the FTP session inbound Active or Passive.

can you pls help on this

[rick jones]

beforeafter - ftp://ftp.cup.hp.com/dist/networking/tools/ is a very useful tool to subtract one set of netstat statistics from another...

Anyhow, look at the Node B before sending to Node A statistics. Notice the TCP retransmission stats:

4287 data packets (5990642 bytes) retransmitted

[rick jones]

beforeafter - ftp://ftp.cup.hp.com/dist/networking/tools/ is a very useful tool to subtract one set of netstat statistics from another...

Anyhow, look at the Node B before sending to Node A statistics. Notice the TCP retransmission stats:

4287 data packets (5990642 bytes) retransmitted
2087 retransmit timeouts

[Rita C Workman]

I may be way off on this one, but....

You mention the speed of 100/Full confirmed. But nowhere above do I see what the card is. On some older lancards we used to have when we were on 10.2, you had to set them to 100/Full Auto_OFF.

So, on the off chance that you are running some old lancard - could you tell us the card model number and the platform O/S you're running?

Thanks,
Rita

[rick jones]

well, twice I've tried to show the after stats, and twice they've been truncated. I've also tried to include how Node A is full duplex autoneg *off* and how unless the switch port is also autoneg off, that will lead to duplex mismatches.

lanadmin -g mibstats 0 should be run on node A to check the link-layer *statistics* and not just the settings - I suspect they will show errors. I suspect checking the switch port stats will also show errors. I won't attach my boilerplate on autoneg a third time :)

[Dennis Handly]

>rick: twice I've tried to show the after stats, and twice they've been truncated.

You're talking about the text in your reply and not the attachment? (You may need to go to the dark side and use a .txt suffix. Though I see your attachment fine.)

[trilok_pvp]

Hi All,

Earlier, Its working fine. 3 weeks back server rebooted. After this reboot i am facing the ftp issue.

shall i change the Auto Neg ON, or it will effect any other. Beacuse its a Production server. I am handling it from remote.

[rick jones]

Autonegotiation *MUST* match on both sides of the link. So, if the server is set explicity to full duplex, the switch must also be set to full duplex. Or, both sides set to auto negotiation. The attachment shows the *statistics* which help show whether or not there is a duplex mismatch between the server and the switch.

[trilok_pvp]

Hi rick,

I am facing the problem when taking backup.
Pls suggest how to solve the problem. I am new to HP UX. I am supporting from remote.

[rick jones]

Please post the full output of "lanadmin -g mibstats 0" from Node A.

[trilok_pvp]

Hi rick,

Pls find the attachement for the NODE A.

Thanks for you support