Information on XP12000 cache algorythm

[Jean-Baptiste Broccard]

Thanks for the comprehensive answer :)
Do you have an idea on the latency metrics ?

Let me give an example:
DG_data_n1 -> lun1 + lun2 + lun3 + lun4
Between date1 and date2 associated average latency on each lun:
lun1 = 0.3ms
lun2 = 0.4ms
lun3 = 0.3ms
lun4 = 0.4ms

Total latency is 1.4ms
Average latency is 0.35ms

Which one do you think is relevant ?
Besides, what is the threshold on this metric I should pay attention (0.1ms, 1ms, 10ms, 100ms) ?

Thanks,
JB

[Peter Mattei]

If I was you I would care not too much on the individual metrics as long as your array performs well.
I would rather use the "Report" function of PA by generating a report that gives you a good performance overview of your array.
Keep it and if you encounter any performance issues you will be able to compare.

Cheers
XP-Pete

PS: From an Server/OS/appliaction standpoint responsetimes <10ms are OK

[Jean-Baptiste Broccard]

Hi Pete,
The reason why I take care of all these metrics is because I don't like using PA for charts: You cannot save the charts, drawing a chart when you already have 10GB of SolidDB is so slow even w/ a 8-core w/ 32GB (!!!)
I don't like drawing report: you cannot choose a group of disks and an accurate timeframe.
So I rewrote a tool that extract the data from an intelligent Db (oracle/mysql, Solid as it's used is *dumb*), create IO profiles mattrix per DG and draw my own charts on a given timeframe.

I now use a xp24k, do you know whether the WP threshold is lower (like 40%) ? Because WP never reach more than 40%, just wondering if it was related to new storage... ?

JB

[Peter Mattei]

Congratulations to the XP24000. This is a real great box!
It is much faster partially due to faster HW but also due to enhanced firmware and algorythms.
So I guess your current load just do not really stress the XP24k and it has plenty of time to destage.

Cheers
XP-Pete