Hi,
Is anybody on this list using FlexCache for NFS access over WAN? What kind of performance metrics would you be tracking to detect potential performance issues? Are you leveraging things like "flexcache hist" for any kind of troubleshooting?
Thanks, Gregory --------------------------------------------------------------------- Intel Israel (74) Limited
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
We do this extensively but with the following limits. 1. Flexcache across the wan volumes are set to VeryLow/VeryLow priority settings. The WAN already slows down access. However, if you let a FC volume have the same priority setting as a Medium/Medium local volume you can see performance impacts to the local volume. The filer tires to give all volumes with the same priority settings similar priorities. If something that is waiting on the WAN to respond starts backing up it's requests sooner or later the filer will try to give most of it's resources to the WAN limited FC volumes. 2. FC volumes are for interactive updates ONLY. We don't allow our users to run LSF jobs against a FC volume. That's not to say they don't do it but we show up with pitchforks when they do. In the case where they need low remote write access but high read access for batch jobs then we mount the FC volume under a different name with write access and we snapmirror the entire file system to the local site. The LSF jobs mount the snapmirror for the heavy lifting read activity.
With those two policies adhered to we rarely see performance impacts. Abuse or ignorance is where we see huge problems. Chris
On 12/04/13 07:06, Touretsky, Gregory wrote:
Hi,
Is anybody on this list using FlexCache for NFS access over WAN?
What kind of performance metrics would you be tracking to detect potential performance issues?
Are you leveraging things like “flexcache hist” for any kind of troubleshooting?
Thanks,
Gregory
Intel Israel (74) Limited
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
Interesting, thank you. Are you using it for high latency links also (>150ms)?
With interactive work, if a user is trying to load some large file which is not in the cache yet - he would be suffering. With batch, even if the first few jobs are taking longer time due to cache population - subsequent jobs would find data locally and run without major performance impact. So we actually see LSF-like usage as one of the accepted scenarios - as long as batch jobs are not trying to write back large amounts of data via FC volumes.
Are you tracking any specific performance indicators - whether at FC volume level, or at the filer level?
-----Original Message----- From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Blackmor, Chris Sent: Thursday, December 05, 2013 05:30 To: toasters@teaparty.net Subject: Re: FlexCache monitoring
We do this extensively but with the following limits. 1. Flexcache across the wan volumes are set to VeryLow/VeryLow priority settings. The WAN already slows down access. However, if you let a FC volume have the same priority setting as a Medium/Medium local volume you can see performance impacts to the local volume. The filer tires to give all volumes with the same priority settings similar priorities. If something that is waiting on the WAN to respond starts backing up it's requests sooner or later the filer will try to give most of it's resources to the WAN limited FC volumes. 2. FC volumes are for interactive updates ONLY. We don't allow our users to run LSF jobs against a FC volume. That's not to say they don't do it but we show up with pitchforks when they do. In the case where they need low remote write access but high read access for batch jobs then we mount the FC volume under a different name with write access and we snapmirror the entire file system to the local site. The LSF jobs mount the snapmirror for the heavy lifting read activity.
With those two policies adhered to we rarely see performance impacts. Abuse or ignorance is where we see huge problems. Chris
On 12/04/13 07:06, Touretsky, Gregory wrote:
Hi,
Is anybody on this list using FlexCache for NFS access over WAN?
What kind of performance metrics would you be tracking to detect potential performance issues?
Are you leveraging things like "flexcache hist" for any kind of troubleshooting?
Thanks,
Gregory
Intel Israel (74) Limited
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
_______________________________________________ Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters --------------------------------------------------------------------- Intel Israel (74) Limited
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
-
Blackmor, Chris -
Touretsky, Gregory