Comware Based
cancel
Showing results for 
Search instead for 
Did you mean: 

A5800 switches, NC522SFP+ 10 Gbps NICs configuration of Jumbo Framing and TCP Chimney

 
Michael A. McKenney
Respected Contributor

A5800 switches, NC522SFP+ 10 Gbps NICs configuration of Jumbo Framing and TCP Chimney

I  configured a A5820 and two A5800 switches in IRF.   I have a DL380 G5 running Windows 2003 32-bit.  I have a DL380 G5 and ML350 G5 running Windows 2003 R2 64-bit.  All three have SP2 and NC522SFP 10 Gbps NIC installed connected to ports on A5820 in the same VLAN.     Each server has a pair of E5420 2.5 GHz quad cores installed.  The two 64-bit servers have 32GB of RAM.   32-bit has 4 GB of RAM.  Only one port is active on each NC522SFP adapter. 

 

 I get under 750-900 Mbps on the 10 Gbps NICs. TCP Chimney is disabled.   Would enabling TCP Chimney or Jumbo Framing increase performance?   What is the best way of testing bandwidth on these adapters? 

2 REPLIES
manuel.bitzi
Trusted Contributor

Re: A5800 switches, NC522SFP+ 10 Gbps NICs configuration of Jumbo Framing and TCP Chimney

Hi Michael

 

I built a similar construct a year ago. HP 7500, DL380G5, NC522SFP, Win2008R2. But reached the same performance as you. I discussed this several times with our serverteam, but now improvments.

 

A big test from our storage partner discovered that Jumbo Frames only have an 7% advantage with hugh data-traffic like iSCSI. Under usual circumstances, jumbo frame is waste of time.

 

br

Manuel

H3CSE, MASE Network Infrastructure [2011], Switzerland
Michael A. McKenney
Respected Contributor

Re: A5800 switches, NC522SFP+ 10 Gbps NICs configuration of Jumbo Framing and TCP Chimney

I did more research on it over the weekend.   I found the same conclusions.