getattr miss
Subsystems (2)
Forgive me for what may be an obvious question. In the middle of tracking down NFS4ERR_RESOURCE storms from an AIX server to linux clients on three busy webservers, I noticed that our linux client performs a compound OPEN,GETATTR for every read of a file even if the reads are well within the attribute cache timeout. Adding nocto doesn't seem to change this behavior -- and for apache looking up numerous .htaccess files it would be great to avoid the trip to the server to revalidate attributes if we're within the timeout.
So the simple question: is this expected? I'd like to minimize consecutive GETATTR calls, if possible, and get the most from VFS cache.
I've got some ridiculous timeouts set up at this point:
barasinga.uvm.edu:/ on /fs/barasinga type nfs4 rw,sec=krb5,nocto,hard,intr,actimeo=3600,addr=132.198.101.73
Ben
Signature · 4 lines
Benjamin Coddington Systems Architecture and Administration Enterprise Technology Services University of Vermont
Quoted 8 lines · Benjamin Coddington
Forgive me for what may be an obvious question. In the middle of tracking down
NFS4ERR_RESOURCEstorms from an AIX server to linux clients on three busy webservers, I noticed that our linux client performs a compound OPEN,GETATTR for every read of a file even if the reads are well within the attribute cache timeout. Adding nocto doesn't seem to change this behavior -- and for apache looking up numerous .htaccess files it would be great to avoid the trip to the server to revalidate attributes if we're within the timeout.
The NFSv4 client has to send an open to the server on each local open.
The one way to avoid this is with delegations. I don't know what sort of delegation support the AIX server has. Probably at least some.
NFSv4.0 delegations require the server to open a separate connection back to the client for recalls; a NAT or firewall could be getting in the way of that.
--b.
Quoted 19 lines
So the simple question: is this expected? I'd like to minimize consecutive GETATTR calls, if possible, and get the most from VFS cache.
I've got some ridiculous timeouts set up at this point:
barasinga.uvm.edu:/ on /fs/barasinga type nfs4 rw,sec=krb5,nocto,hard,intr,actimeo=3600,addr=132.198.101.73
Ben
Signature · 8 lines
Benjamin Coddington Systems Architecture and Administration Enterprise Technology Services University of Vermont -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Looks like our AIX server supports read delegations, but they don't seem to be used. NAT/Firewall is not an issue here. Instead, the AIX server always requires OPEN_CONFIRM, and so there are no delegations. So now I'm trying to figure out why every open needs a confirm.. Thanks.
B
Quoted 31 lines · J. Bruce Fields
Quoted 1 line · Benjamin Coddington
Forgive me for what may be an obvious question. In the middle of tracking down
NFS4ERR_RESOURCEstorms from an AIX server to linux clients on three busy webservers, I noticed that our linux client performs a compound OPEN,GETATTR for every read of a file even if the reads are well within the attribute cache timeout. Adding nocto doesn't seem to change this behavior -- and for apache looking up numerous .htaccess files it would be great to avoid the trip to the server to revalidate attributes if we're within the timeout.The NFSv4 client has to send an open to the server on each local open.
The one way to avoid this is with delegations. I don't know what sort of delegation support the AIX server has. Probably at least some.
NFSv4.0 delegations require the server to open a separate connection back to the client for recalls; a NAT or firewall could be getting in the way of that.
--b.
Quoted 16 lines
So the simple question: is this expected? I'd like to minimize consecutive GETATTR calls, if possible, and get the most from VFS cache.
I've got some ridiculous timeouts set up at this point:
barasinga.uvm.edu:/ on /fs/barasinga type nfs4 rw,sec=krb5,nocto,hard,intr,actimeo=3600,addr=132.198.101.73
Ben
Signature · 8 lines
Benjamin Coddington Systems Architecture and Administration Enterprise Technology Services University of Vermont -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html