Performance statistics?
Performance statistics?
So the performance statistics are instant (from the point you open the performance box) and not historic (from the time FancyCache is launched)????
Re: Performance statistics?
Yes, currently it is.
-
- Level 1
- Posts: 3
- Joined: Fri Sep 17, 2010 3:34 pm
Re: Performance statistics?
Having problems uploading to this forum. Will post back later.
-
- Level 1
- Posts: 3
- Joined: Wed Nov 03, 2010 10:58 pm
Re: Performance statistics?
Can you please explain to me whats the difference of "Write Bytes (Total)", "Write Bytes (Deferred, Urgent)" and "Write Bytes (Defered, Normal)"?
And What about "Waited IOs".
And What about "Waited IOs".
Re: Performance statistics?
Hi,
Write Bytes (Total): the total bytes received from the IO requests which need to be written to the volume/disk.
Write Bytes (Deferred): the total bytes written to the volume/disk from the cache.
urgent: the defer-written operation is triggered by the insufficient free cache space.
normal: the defer-written operation is triggered by the write latency.
Waited IOs: this is for our internal use.
Write Bytes (Total): the total bytes received from the IO requests which need to be written to the volume/disk.
Write Bytes (Deferred): the total bytes written to the volume/disk from the cache.
urgent: the defer-written operation is triggered by the insufficient free cache space.
normal: the defer-written operation is triggered by the write latency.
Waited IOs: this is for our internal use.
Re: Performance statistics?
Urgent: I understood it can happen earlier (In case there is a problem with cache size.)
Normal: FC tries to write the data as soon as possible but only if disk subsystem is not busy. Con you confirm this?
Recently I've spend some time by researching disk performance issues, counters and metrics... So I've been wondering what methodology do you use to determine if disks are busy or not?
Is it "%idle time", "current disk queue", or "avg. disk sec/transfer"?
thx
--Lukas
Normal: FC tries to write the data as soon as possible but only if disk subsystem is not busy. Con you confirm this?
Recently I've spend some time by researching disk performance issues, counters and metrics... So I've been wondering what methodology do you use to determine if disks are busy or not?
Is it "%idle time", "current disk queue", or "avg. disk sec/transfer"?
thx
--Lukas
Re: Performance statistics?
Hi Lukas,
Not correct. Pls refer to my previous post.
Not correct. Pls refer to my previous post.
We'll update our write policy to make it smoother.horizon wrote:Urgent: I understood it can happen earlier (In case there is a problem with cache size.)
Normal: FC tries to write the data as soon as possible but only if disk subsystem is not busy. Con you confirm this?
Re: Performance statistics?
Well, not sure I follow.
DeferredUrgent write: This occurs if there are too many data waiting write and cache is full. (If I am wrong I cannot imagine the different scenario for DeferredUrgent writes )
DeferredNormal: You said I was not correct. This would than mean that in current version the writing are just statically posponed for fixed amount of time no matter what. (In my case 5 minutes.)
So it can happen that as soon as the fixed amount of time passes, disk can be busy (by some other activity) again.
This would mean that DeferredNormal can collide with another disk activity and cause the performance problem. :-O
Well, I believe that DeferredNormal should smoothly "rain" data to the disk. (=Maintaining that PC (=disk subsystem + cpu? + ...?) is still relaxed and all other requests (cpu, disk reads, ...) can be satisfied without any delay.
--Lukas
DeferredUrgent write: This occurs if there are too many data waiting write and cache is full. (If I am wrong I cannot imagine the different scenario for DeferredUrgent writes )
DeferredNormal: You said I was not correct. This would than mean that in current version the writing are just statically posponed for fixed amount of time no matter what. (In my case 5 minutes.)
So it can happen that as soon as the fixed amount of time passes, disk can be busy (by some other activity) again.
This would mean that DeferredNormal can collide with another disk activity and cause the performance problem. :-O
Well, I believe that DeferredNormal should smoothly "rain" data to the disk. (=Maintaining that PC (=disk subsystem + cpu? + ...?) is still relaxed and all other requests (cpu, disk reads, ...) can be satisfied without any delay.
--Lukas
Re: Performance statistics?
Hi Lukas,
yes, you are right.
yes, you are right.