vBulletin Search Engine Optimization
| |||||||
| Register | FAQ | Members List | Calendar | Search | Today's Posts | Mark Forums Read |
| ||||
| Hi All, We've been seeing some weirdness on one of our nfs clusters with gigabit interfaces for the public network connections. The machines are 4500s (so we have sbug ge cards) on solaris 8 108528-18. Under a certain amount of load, we begin to see input errors accumulate a significant ratio to successful packets reflected via netstat -i on the nfs server. We dont see any evidence of the issue on our switches. The curious bit is, we've changed cables/switch ports/etc. with no luck. If i fail over the resource group to another node in the cluster - the errors follow the load. I'm wondering if it could be a bogus nfs client (all clients are running 108528-18 as well). Average throughput on the interfaces is about 5 mb/sec when this begins to occur, but can be even lower. When there is much more limited nfs traffic, the errors don't occur at all. Any ideas? tia - justin -- Justin |
| |||
| Justin Shaffer <shafferj@mac.com> wrote in message news:<250920031509103588%shafferj@mac.com>... > Hi All, > > We've been seeing some weirdness on one of our nfs clusters with > gigabit interfaces for the public network connections. The machines are > 4500s (so we have sbug ge cards) on solaris 8 108528-18. Under a > certain amount of load, we begin to see input errors accumulate a > significant ratio to successful packets reflected via netstat -i on the > nfs server. We dont see any evidence of the issue on our switches. The > curious bit is, we've changed cables/switch ports/etc. with no luck. If > i fail over the resource group to another node in the cluster - the > errors follow the load. I'm wondering if it could be a bogus nfs client > (all clients are running 108528-18 as well). Average throughput on the > interfaces is about 5 mb/sec when this begins to occur, but can be even > lower. When there is much more limited nfs traffic, the errors don't > occur at all. Any ideas? > > tia - justin Hey, is this sun cluster 3.0? You have many NFS-Directorys? Try ha-nfs+ (plus), a SUN addon to the cluster, it is for heavy loaded NFS-Services. CU Klaus Grote |
| |||
| Justin Shaffer <shafferj@mac.com> wrote: > Hi All, > We've been seeing some weirdness on one of our nfs clusters with > gigabit interfaces for the public network connections. The machines are > 4500s (so we have sbug ge cards) on solaris 8 108528-18. Under a > certain amount of load, we begin to see input errors accumulate a > significant ratio to successful packets reflected via netstat -i on the > nfs server. We dont see any evidence of the issue on our switches. What does netstat -k <interface> show? rick jones -- a wide gulf separates "what if" from "if only" these opinions are mine, all mine; HP might not want them anyway... feel free to post, OR email to raj in cup.hp.com but NOT BOTH... |
| ||||
| Sooooo... here is whatwe found. Thanks for the netstat -k recommendation - i dont know why that didnt occur to us before. Basically, when ge_queue_cnt goes up - we see the ge_queue_full_cnt increment. After analysing the traffic a bit more, it looks just like a rate issue on the interface - not a bandwidth issue - due to the large number of stats coming across the interface. Regards, Justin In article <So_cb.5789$nT3.5692@news.cpqcorp.net>, Rick Jones <foo@bar.baz.invalid> wrote: > Justin Shaffer <shafferj@mac.com> wrote: > > Hi All, > > > We've been seeing some weirdness on one of our nfs clusters with > > gigabit interfaces for the public network connections. The machines are > > 4500s (so we have sbug ge cards) on solaris 8 108528-18. Under a > > certain amount of load, we begin to see input errors accumulate a > > significant ratio to successful packets reflected via netstat -i on the > > nfs server. We dont see any evidence of the issue on our switches. > > What does netstat -k <interface> show? > > rick jones -- Justin Shaffer |
| Thread Tools | |
| Display Modes | |
|
|