I have an HA pair, both with pam. One does NOT have PAM enabled. We have some uniquely bursty workloads at the top of every hour, and they are unusually high peaks 3 or 4 hours every day in the AM.
The PAM enabled system suffers VERY High concurrency & latency on these few peaks, but the non PAM system does not.
PAM hit rate is about 60%.
I'm wondering if this is because these sudden user-based spike workloads are almost 0% cachable, that Kahuna takes a hit evicting an unusually high amount of blocks from PAM..and updating the in-ram hash of what's actually in PAM. When I enable PAM on controller 2, this odd concurrency/latency spike then also shows up on that head too.
Thoughts? _________________________________Jeff MohlerTech Yahoo, Storage Architect, Principal(831)454-6712TW: @PrincipalYahooYM: Supra89ta
Is it really pam, or pam II or flexscale cache?
What are the options flexscale currently set to?
From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Jeffrey Mohler Sent: Wednesday, April 15, 2015 1:34 PM To: toasters@teaparty.net Subject: Odd Kahuna behaviour W/Pam
I have an HA pair, both with pam.
One does NOT have PAM enabled.
We have some uniquely bursty workloads at the top of every hour, and they are unusually high peaks 3 or 4 hours every day in the AM.
The PAM enabled system suffers VERY High concurrency & latency on these few peaks, but the non PAM system does not.
PAM hit rate is about 60%.
I'm wondering if this is because these sudden user-based spike workloads are almost 0% cachable, that Kahuna takes a hit evicting an unusually high amount of blocks from PAM..and updating the in-ram hash of what's actually in PAM.
When I enable PAM on controller 2, this odd concurrency/latency spike then also shows up on that head too.
Thoughts?
_________________________________ Jeff Mohlermailto:jmohler@yahoo-inc.com Tech Yahoo, Storage Architect, Principal (831)454-6712 TW: @PrincipalYahoo YM: Supra89ta
Did you check for proper slot assignment? If it is not in the proper slot, it could theoretically cause slowness
--tmac
*Tim McCarthy* *Principal Consultant*
On Wed, Apr 15, 2015 at 1:34 PM, Jeffrey Mohler jmohler@yahoo-inc.com wrote:
I have an HA pair, both with pam.
One does NOT have PAM enabled.
We have some uniquely bursty workloads at the top of every hour, and they are unusually high peaks 3 or 4 hours every day in the AM.
The PAM enabled system suffers VERY High concurrency & latency on these few peaks, but the non PAM system does not.
PAM hit rate is about 60%.
I'm wondering if this is because these sudden user-based spike workloads are almost 0% cachable, that Kahuna takes a hit evicting an unusually high amount of blocks from PAM..and updating the in-ram hash of what's actually in PAM.
When I enable PAM on controller 2, this odd concurrency/latency spike then also shows up on that head too.
Thoughts?
Jeff Mohler Tech Yahoo, Storage Architect, Principal (831)454-6712 TW: @PrincipalYahoo YM: Supra89ta
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
-
Jeffrey Mohler -
Jordan Slingerland -
tmac