How are the NICs on the F520 configured? How are the switch ports configured?
Make sure that the duplexes match on both ends.
-Devi
At 04:09 PM 5/14/98 -0500, Brett Rabe wrote:
Most of ya'll probably understand what it feels like to be frustratingly baffled, so pull up a chair, lend me an ear, and I'll buy ya a beer.
I've got two NetApps (and two more on order).
- F230
o 5.0R2 OnTap o 128MB RAM o 4MB NVRAM o 5 9-gig drives o 2 100bT NIC
- F520
o 5.0R2 OnTap o 192MB RAM o 4 MB NVRAM o 9 9-gig drives o 2 100bT NIC
The installation is as follows. Both NetApps are fronted by three FreeBSD servers. One NIC on the NetApps routes to its own private subnet. Each of the three servers have a second NIC card also routing to that subnet.
I can't get squat for performance from the F520. I'm not trying to pretend that using 'yes' is a true benchmark, but it does illustrate the point sufficiently:
F520 front-end# yes > /some_nfs_mounted_filesystem/yes F520> sysstat 1 CPU NFS CIFS HTTP Net kB/s Disk kB/s Tape kB/s Cache in out read write read write age 2% 35 0 0 445 8 0 0 0 0 >60 4% 36 0 0 468 9 0 0 0 0 >60 18% 37 0 0 469 9 1684 816 0 0 >60 10% 30 0 0 377 7 96 1632 0 0 >60 15% 157 0 0 1457 33 0 0 0 0 >60 29% 84 0 0 870 18 1336 2348 0 0 >60
F230 front-end# yes > /some_nfs_mounted_filesystem/yes F230> sysstat 1 CPU NFS CIFS HTTP Net kB/s Disk kB/s Tape kB/s Cache in out read write read write age 68% 287 0 0 1510 823 1052 2300 0 0 >60 63% 212 0 0 1683 751 24 1628 0 0 >60 67% 248 0 0 1678 811 944 1960 0 0 >60 59% 248 0 0 1680 780 1068 1632 0 0 >60 59% 214 0 0 1727 804 1048 1760 0 0 >60 60% 215 0 0 1684 805 43 2285 0 0 >60 53% 202 0 0 1741 783 1052 2328 0 0 >60 57% 179 0 0 1416 758 1064 2320 0 0 >60
(The F230 also was being backed up and is in production hosting web pages, in case anyone notes about all the outbound traffic that wouldn't be accounted for by 'yes'....)
The NetApps are configured as close to identical as reasonable. The version of FreeBSD appears to be not a factor -- I've got both 2.2.5-RELEASE and 2.2.6-RELEASE that produce the second results on the F230, and 2.2.5-RELEASE and 2.2.6-RELEASE that produce the first results on the F520.
All the interfaces are configured correctly. There are no significant errors reported by the NetApps, the FreeBSD boxes, or the network device to which they all terminate. I can't discern any significant difference in the FreeBSD box configurations -- if anything, the fact that I get the same performance with different FreeBSD OS revs on the F230 and the same poor performance with different FreeBSD OS revs on the F520 points to a problem with the F520.
I had the F230 at 4.3.1 and got similar performance as I do with 5.0R2. I've gone so far as to swap ethernet cables. :p
I'm not sure where to look next; suggestions would be highly welcomed. I have (of course) hit now.netapp.com and hunted through their troubleshooting suggestions to no avail.
... just another day. :)
Brett
Brett Rabe Email : brett@uswest.net Systems Administrator - U S West Phone : 612.664.3078 Interact - 3S Pager : 612.613.2549 600 Stinson Blvd. Pager : page-brett@uswest.net Minneapolis, MN USA Fax : 612.664.4770
All those who believe in psychokinesis, raise my hand!
Well, yes is probably not the best test for pumping through the data either.
Try using something like IOZONE, mount the drives UDP/v2, and let it rock.
I regularly see 8-9MB/second on mine over FastEthernet.
Of course, the obvious things like checking collisions/full-half-duplex/ settings and all that stuff apply.
On Thu, 14 May 1998, Devi Nagaraj wrote:
How are the NICs on the F520 configured? How are the switch ports configured?
Make sure that the duplexes match on both ends.
-Devi
At 04:09 PM 5/14/98 -0500, Brett Rabe wrote:
Most of ya'll probably understand what it feels like to be frustratingly baffled, so pull up a chair, lend me an ear, and I'll buy ya a beer.
I've got two NetApps (and two more on order).
- F230
o 5.0R2 OnTap o 128MB RAM o 4MB NVRAM o 5 9-gig drives o 2 100bT NIC
- F520
o 5.0R2 OnTap o 192MB RAM o 4 MB NVRAM o 9 9-gig drives o 2 100bT NIC
The installation is as follows. Both NetApps are fronted by three FreeBSD servers. One NIC on the NetApps routes to its own private subnet. Each of the three servers have a second NIC card also routing to that subnet.
I can't get squat for performance from the F520. I'm not trying to pretend that using 'yes' is a true benchmark, but it does illustrate the point sufficiently:
F520 front-end# yes > /some_nfs_mounted_filesystem/yes F520> sysstat 1 CPU NFS CIFS HTTP Net kB/s Disk kB/s Tape kB/s Cache in out read write read write age 2% 35 0 0 445 8 0 0 0 0 >60 4% 36 0 0 468 9 0 0 0 0 >60 18% 37 0 0 469 9 1684 816 0 0 >60 10% 30 0 0 377 7 96 1632 0 0 >60 15% 157 0 0 1457 33 0 0 0 0 >60 29% 84 0 0 870 18 1336 2348 0 0 >60
F230 front-end# yes > /some_nfs_mounted_filesystem/yes F230> sysstat 1 CPU NFS CIFS HTTP Net kB/s Disk kB/s Tape kB/s Cache in out read write read write age 68% 287 0 0 1510 823 1052 2300 0 0 >60 63% 212 0 0 1683 751 24 1628 0 0 >60 67% 248 0 0 1678 811 944 1960 0 0 >60 59% 248 0 0 1680 780 1068 1632 0 0 >60 59% 214 0 0 1727 804 1048 1760 0 0 >60 60% 215 0 0 1684 805 43 2285 0 0 >60 53% 202 0 0 1741 783 1052 2328 0 0 >60 57% 179 0 0 1416 758 1064 2320 0 0 >60
(The F230 also was being backed up and is in production hosting web pages, in case anyone notes about all the outbound traffic that wouldn't be accounted for by 'yes'....)
The NetApps are configured as close to identical as reasonable. The version of FreeBSD appears to be not a factor -- I've got both 2.2.5-RELEASE and 2.2.6-RELEASE that produce the second results on the F230, and 2.2.5-RELEASE and 2.2.6-RELEASE that produce the first results on the F520.
All the interfaces are configured correctly. There are no significant errors reported by the NetApps, the FreeBSD boxes, or the network device to which they all terminate. I can't discern any significant difference in the FreeBSD box configurations -- if anything, the fact that I get the same performance with different FreeBSD OS revs on the F230 and the same poor performance with different FreeBSD OS revs on the F520 points to a problem with the F520.
I had the F230 at 4.3.1 and got similar performance as I do with 5.0R2. I've gone so far as to swap ethernet cables. :p
I'm not sure where to look next; suggestions would be highly welcomed. I have (of course) hit now.netapp.com and hunted through their troubleshooting suggestions to no avail.
... just another day. :)
Brett
Brett Rabe Email : brett@uswest.net Systems Administrator - U S West Phone : 612.664.3078 Interact - 3S Pager : 612.613.2549 600 Stinson Blvd. Pager : page-brett@uswest.net Minneapolis, MN USA Fax : 612.664.4770
All those who believe in psychokinesis, raise my hand!
Well, yes is probably not the best test for pumping through the data either.
What, by the way, is "yes"?
Try using something like IOZONE, mount the drives UDP/v2, and let it rock.
Why do you suggest UDP/v2, and not UDP/v3?
Does FreeBSD support >8KB V3 transfer sizes yet?
I regularly see 8-9MB/second on mine over FastEthernet.
On write?
Of course, the obvious things like checking collisions/full-half-duplex/ settings and all that stuff apply.
I just got a dump of all this from the fellow who started this thread, stay tuned.
What, by the way, is "yes"?
YES(1) FreeBSD General Commands Manual YES(1)
NAME yes - be repetitively affirmative
SYNOPSIS yes [expletive]
DESCRIPTION Yes outputs expletive, or, by default, ``y'', forever.
HISTORY The yes command appeared in 4.0BSD.
4th Berkeley Distribution June 6, 1993 1
-- Karl Swartz - Technical Marketing Engineer Network Appliance Work: kls@netapp.com http://www.netapp.com/ Home: kls@chicago.com http://www.chicago.com/~kls/
On Thu, 14 May 1998, Brian Pawlowski wrote:
Try using something like IOZONE, mount the drives UDP/v2, and let it rock.
Why do you suggest UDP/v2, and not UDP/v3?
IN all my tests, with FreeBSD clients, v3 has never beat v2, and since in my cases the nodes are on a private non-bridged/non-routed subnet, the utility of TCP vs UDP in this case seems lost.
Does FreeBSD support >8KB V3 transfer sizes yet?
I regularly see 8-9MB/second on mine over FastEthernet.
On write?
Writes are 6-7, Reads 8-9. I don't know about >8kb transfer sizes as I have no desire to mess with a perfectly working stable system... :)
Try using something like IOZONE, mount the drives UDP/v2, and let it rock.
Why do you suggest UDP/v2, and not UDP/v3?
IN all my tests, with FreeBSD clients, v3 has never beat v2, and since in my cases the nodes are on a private non-bridged/non-routed subnet, the utility of TCP vs UDP in this case seems lost.
TCP is not required for V3 - it works as well with V2 (only always more slowly through any combo of vendor's servers and clients I can muster).