I would like to add a few more details to the problem i'm having with the poor NFS performance on the Filers using NFS over the WAN(See towards the bottom). The remote Solaris system is 2.8 The NetApp filer(740) is 6.2.2 The Filer has a quad board to Etherchannel ports on a Cisco switch 100mb is the backbone on LAN
My first suspicion was maybe the filers were slow on the LAN also but since i was running 100mb I didn't notice the problem. However, this is not the case. I forgot to add i did do extensive testing on the LAN at corporate and I was getting between 6-8.5MBytes/sec from Filer to Solaris system and from Solaris(local disk) to another Solaris(local disk) system. There really was not much difference between the performance on the Filer or the Solaris box. Also, when I used the UDP transport for the remote NFS the performance was terrible. I was seeing a lot of retransmissions. TCP definitely works better on the WAN. Also, the RTT's were the same from the Remote Solaris system to the corporate Filer or from the Remote Solaris system to the corporate Solaris system.
I also want to thank several people for their comments on this subject. Unfortunately, i still have been unable to identify the source of the problem. Jeff Mohler Neil Stichbury Alan McLachlan Jeff Wolfe Hunter Wylie
Mike McBride Silicon Wave, Inc. (858) 404-5210 mcbride@siliconwave.com http://www.siliconwave.com http://www.siliconwave.com/
-----Original Message----- From: Hunter Wylie [mailto:hwylie@stpaultel.com] Sent: Wednesday, April 16, 2003 10:39 PM To: Mike McBride Subject: Re: Slow NFS filer performance on WAN connection vs Solaris NFS WAN connection
Mike, Here's a simple test to try isolate the problem. From a third system ("C") at the same central site run the test talking the Sun ("A") and the 740 ("B"). Run the test on the local LAN and see what you get. If the filer speeds are the same and the Filer is loafing you have a networking problem in the LAN. Try manually setting all network interfaces to Full. Never use auto detection mode. What speed interface is each system, A & B?
If the Filer runs at about 6-8.5mbytes per second, assuming a 100mbit LAN interface on the file you still have network problems somewhere between the filer and the VPN box
The problems you describe are normally found in the network when the filer is loafing. Check the stats on the LAN interfaces on the filer too, lost frames, retries, etc.. Might be a bad Filer LAN interface.
Hunter
----- Original Message ----- From: Mike mailto:mcbride@siliconwave.com McBride To: 'toasters@mathworks.com' mailto:'toasters@mathworks.com' Cc: Mike McBride mailto:mcbride@siliconwave.com Sent: Wednesday, April 16, 2003 4:32 PM Subject: Slow NFS filer performance on WAN connection vs Solaris NFS WAN connection
We have several engineers working from home using Solaris boxes and utilizing VPN's to the office. Typical application use is Cadence. What i have noticed is the NFS performance on the Filers are almost 50% slower than the performance of the Solaris machines. I have run several tests including cp, ftp and utilizing ping roundtrip times, traceroute, snoop and tcpdump to monitor activities between home SUN systems and file systems at work. I have also tested NFS using different transport protocols(tcp/udp) modifying rsize/wsize buffers and using NFS2 and NFS3 versions of NFS. The filers and the Solaris systems at cooperate are on the same subnet - there is no differences in the routing. FYI - the VPN's at home are all Linksys VPN cable/dsl routers configured for Site to Site VPN connections to corporate office.
The problem is definitely with NFS and it is isolated to the Filers. I am confused why the performance of the NetApp NFS would be so poor compared with NFS on the Solaris systems. Now, I know NFS on a WAN is slow and VPN encryption is going to degrade the performance severely, but I am wondering why NetApp NFS performance is much worse than Solaris NFS performance. Basically, when accessing files on a Solaris system(local disk) at the corporate office from home i'm getting about 915KBits/sec. When accessing the same files on a NetApp(F740) at the corporate office from home i'm getting about 510KBits/sec. The transfer mechanism is NFS. When using FTP i get the same 915KBits/sec on the Solaris(local disk) and Filer. The load on the Filer at the time of testing is minimal.
Anyone have any insight on this subject? Any feedback would be appreciated since I put quite a bit of time in this to try and understand the problem here.
Thanks,
Mike McBride Silicon Wave, Inc. (858) 404-5210 mcbride@siliconwave.com http://www.siliconwave.com http://www.siliconwave.com