Ace with servers in VMware

[mikik]

Hi;

I have a customer who has a test site  with one ACE doing load balancing for a small farm ( 8 servers). Recently the customer moved his servers to

to VMware. The customer claims that since that change, the ACE is causing for large delays. His claim is that when he points his clients http requests directly to the servers, bypassing the ACE, he  receives normal response time. His claim is that the ACE is causing up to 30 sec delays. This was definitely not the case before they made their change.

I was wondering if anyone has any insight to this type of situation ? Are there any specific ACE issues and load balancing factors that may surface when working with VMware, which are not notticable with real servers?

The ACE is blade in  6500

Thanks for any help.

Mickey

[Amit Singh · Accepted solution]

I havenot heared of anything like that.. its  a serious 30 sec delay. Did you check the traffic flow after migrating to VM? ideally it should not make any difference.

Check this :

http://www.cisco.com/en/US/partner/docs/solutions/Enterprise/Data_Center//vmware/cisco_VMwareView.html#wp244841

Cheers,

-amit singh

[venkatkr · Accepted solution]

From an ACE point of you, ACE does not care what type or rservers you have. You can run the Rserver on any systems or OS. What you need is a sniffer trace from you customer to confirm what they claim. If ACE is proxying the connection then you need to see in the trace if the delay is with the ACE or with the network (which includes the vmware).

If you need help with that, please post the trace and the configuration.

Thanks

V.K

[mikik]

Hi and thanks for the answers:

Venkatkr, I will be at the customer tomorow and was thinking of placing a sniffer at some point where packets come in and leave the ACE.  Since this ACE is implemented as a module in a 6500, I was thinking of trying to set up a monitor session on the  6500, on the vlan going  to

the module. Hopefully I will see both requests and replies and be able  to see the amount of time  that passes between the two.

Does this sound reasonable or there a better way to set up a trace?

Again thanks for the help.

Mickey

[venkatkr]

Hi Mickey,

Because this is a ACE module, you can just sniff the ten gig interface on the ACE. This way you will get everything coming in and out of the ACE. Now if you have lot of traffic then this will be bit overwhelming.  So may be you can find a lean period and do this exercise or possible pick a client PC from where you can repro the slowness and filter based on that.

Also as you will be using wireshark, you can write to multiple files so that you dont loose the interesting traffic.

I have attached the process of doing a ten gig capture to this post. Hope this helps

Cheers

V.K

[mikik]

Hi:

Thanks for the material. This is what I will do tomorow. It's a small development site , so I think I can filter the connections I need. Will update when have results.

Thanks

[mikik]

Hi:

Just to close the case, the problem boiled down to MTU. The site is configured to work Jumbo frames on most interfaces. The customer forgot to add

MTU to a vlan interface . This caused slow response. Nothing to do with VMware. The ACE works fine with VMware and with Jumbo frames.

Thanks for all the help.

Mickey