The F230's performance is not limited by CPU clock rate. It is limited by the performance of the host bridge (a.k.a. north bridge). The host bridge defines memory bandwidth and PCI bandwidth. Overclocking the F230 won't solve a heavily loaded filer.
David
-----Original Message----- From: Dane Jasper [mailto:dane@sonic.net] Sent: Thursday, February 04, 1999 5:09 PM To: toasters@mathworks.com Subject: NetApp CPU upgrade
Has anyone considered upgrading the CPU on their NetApp product? It's a closed product, so NetApp won't talk about it, but it seems to me that the Pentium 90 in my F230 (66mhz external speed, 90mhz core?) could be replaced with a Pentium 133 with the same 66mhz external speed without any need for jumpers, etc.
My NetApp is in heavy service, so I'm a bit hesitant. Has anyone tried this, or does anyone have a NetApp in light service that they'd like to try this with?
I've searched the listserve archive and don't see any subjects that would seem to have previously addressed this.
Somewhat busy CPU:
CPU NFS CIFS HTTP Net kB/s Disk kB/s Tape kB/s Cache in out read write read write age 24% 431 0 0 74 672 573 0 0 0 2 36% 639 0 0 149 1035 768 0 0 0 2 30% 524 0 0 129 708 639 0 0 0 2 44% 868 0 0 158 925 868 0 0 0 2 42% 857 0 0 179 734 560 0 0 0 2 37% 658 0 0 133 735 1069 0 0 0 2 29% 494 0 0 100 737 752 0 0 0 2 27% 459 0 0 80 834 724 0 0 0 2 44% 886 0 0 189 1078 668 0 0 0 2 53% 717 0 0 151 923 2249 35 0 0 2 86% 589 0 0 104 1050 2678 3423 0 0 2 28% 535 0 0 101 563 212 0 0 0 2 36% 712 0 0 133 654 244 0 0 0 2
Regarding cache age, this unit already has the full 256M of RAM, so we can't address that issue in the conventional way either.
David, is there a way to find out what the load on this host bridge is currently at so that we can project end of life for this installed filer?
Also, does adding a second SCSI controller split up this load, or is it all on the same host bridge? We're running our F230 in a supported but non-standard four shelf configuration with two controllers, but most disks are on the first controller currently.
We're seeking ways to keep this NetApp in useful service as long as possible, and certainly appreciate any feedback you can give on reducing bottlenecks.
We are currently considering a massive increase in the RAM in our web servers which would be used by the kernel for disk buffers. The theory is that if the kernel simply asks the filer if the data file has changed, and it hasn't, and it's served out of disk buffer cache RAM on the web server itself, this would be less load on the filer because the filer never ends up serving the file if there's a cache hit in the web server client.
What are your thoughts on this?
Currently our web servers have 256M of ram, of which only about 32M is used for disk buffers. We're considering increasing this to 1024M of RAM, most of which would be used for buffers. Price for this upgrade to our web servers is about $1800 each, so it's a lot less painful than a filer upgrade.
Thanks for the feedback.
-Dane
On Fri, Feb 05, 1999 at 08:28:23AM -0800, Bulfer, David wrote:
The F230's performance is not limited by CPU clock rate. It is limited by the performance of the host bridge (a.k.a. north bridge). The host bridge defines memory bandwidth and PCI bandwidth. Overclocking the F230 won't solve a heavily loaded filer.
David