massive issue with hpux on hp blades

[carl_townshend]

Hi all

We have an ongoing issue of really slow file transfers to our hpux servers running on blade chassis using virtual connect.

The transfer download starts off around 3Mbs then goes to 512k, however the upload is much quicker. We are transferring a 75Mbit file. we have done a windows copy to a pc, and also via ftp and its the same result.

Now...here is the strange thing, from the local LAN via vlan routing the speeds are fine. However if we go to any of our sites sitting on 100Mbit ethernet lan extension circuits we get the speeds above. There is no issues at all with the WAN as we have checked these and all other applications / servers coming over the WAN is fine, and its nearly the same setup as our LAN to be honest. Its purely to do with these servers running on this enclosure.

We have tried the following

Fixed port speed to 1000M full from our 3750 to the hp virtual connect switch

rebooted the servers

rebooted the switch

had a health check on the servers

updated firmware on the servers

bypassed the 3750 and plugged into another switch

We have done a packet capture and it seems as the client will request the data, then there will be a 1 second pause, then lots of data, this seems to be repeated over and over again. lots of duplicate ACKS

We have ran out of ideas,

Has anyone seen this before or are there any known issues with file transfer / slow network speeds going to hpux servers running on virtual connect ?

really could do with some help with this

cheers

Carl

[Jan Rolny]

Hi Carl,

by my opinion problem would be close to VirtualConnect. Have you checked VirtualConnect settings and assigned speed to particular NIC(in HP terminology LOM interface). Also check VNET settings. Not sure if VirtualConnect can do QoS based operations.

Check uplink configuration (where VC is connect to switch) on your Switches. What type of uplink do you use? Is it optic fiber or standart RJ45 metalic uplink? Do you use LACP?

I have another issue with VirtualConnect but it is know issue with MAC flapping. HP says about LLDP protocol which is used with VC but not by Cisco.

Just suggestion where would be problem.

Best regards,

Jan

[carl_townshend]

hi

I know the server guys have fixed the speed and duplex on the virtual connect side

My question is, what about the HPUX servers internal NIC that connects to the virtual connect ?

what is the LOM interface? are these the ones that show on virtual connect ? or on the actual hpux servers ?

yes we use LACP and yes we do get the mac flaps, but is that right ?

why do we nmeed to enabled LLDP ?

[Jan Rolny]

Hi,

if you connect ot VirtualConnect WEB GUI and will go deep to configuration of servers so you will see so-called LOM adapters. These LOM adapters are real physical NIC's on Blade server. BUT one LOM adapter you can divide to four Virtual NIC's. This VNIC's are visible in system(Linux/windows).

I think LOM is visible just for VC and VNIC is visible for both VC and OS. In VC you can configure VirtualNet and this VirtualNet is connected via one of uplink of VC to real switch.

LACP configuration is definitely right but depends on both side configuration (VC vs. cisco) and requirements.

MAC FLAP is definitely not right and need to be solved. If MAC flap occurs in you network so STP will probably have heavy load and can cause network loops and problems.

I am not sure why LLDP is required to be enabled on cisco switch. LLDP is Link Layer Discovery Protocol and is simillar to CDP. LLDP is worldwide standard CDP is proprietary and used by Cisco.

Regards,

Jan