Re: EVA-5000 performance monitoring
From: Jilly (jilly_at_clarityconnect.com)
Date: 06/20/03
- Next message: Tom Linden: "RE: Why repost this? (was Re: [OT] Humour)"
- Previous message: fernando.vallarino_at_oca.com.uy: "Re: Problems with SYS$GETRMI"
- In reply to: jlsue: "Re: EVA-5000 performance monitoring"
- Next in thread: Hal Kuff: "Re: EVA-5000 performance monitoring"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Date: Fri, 20 Jun 2003 13:08:17 -0400
In my work, ideally, I like to see the following for disk IO statistics
>From a process level, per file stats
Average IO size for reads
Average IO size for writes
Average read rates
Average write rates
Average read response time
Average write response time
Average XFC cache hit %
>From a system level, per volume stats
Average IO size for reads
Average IO size for writes
Average read rate
Average XFC hit %
Average write rate
Average read response time (non-XFC)
Average write response time
>From a storage level, per spindle stats
Average IO size for reads
Average IO size for writes
Average read rate
Average write rate
Average read response time
Average write response time
Haven't found any data collector yet that will supply all of these
items. But having all of these would make it oh so much easier to
troubleshoot performance problems and to do capacity planning.
jlsue wrote:
>
> On Wed, 18 Jun 2003 07:31:49 -0400, "Hal Kuff" <kuff@tessco.com> wrote:
>
> > We'll look into that.... I think it does not show I./O at the unit
> >level... ie.e $1$Dga2130: .... I'm told that EMC has something that can get
> >stats from an HP SAN, but HP does not.... how could that be?
> >
>
> Wait a sec.... $1$DGA2130 is a host-level device unit. If in a cluster,
> you could use a cluster-wide data collection tool that combines each
> system's stats into a complete picture. If not a cluster, then monitor
> disk/item=all would give you the I/O stats to that disk.
>
> I often wonder what people really need to see on the I/O stats at the
> controller level. Nobody ever asked for it for disks on DSSI controllers,
> or SCSI controllers.
>
> Instead, we managed performance by watching I/O queues on the devices and,
> when they were starting to grow on a consistent basis, then we'd address
> the issue (actually, prior to that we'd be doing capacity planning). Sure,
> I've run VDTPY on occasion on the HSC/Z/G, but it's never been a real
> integral part of my performance management and capacity planning (and I've
> done a LOT of that over the years). With these HS controllers, using SCSI
> technology, there is a relatively small number of spindles involved in each
> LUN, and performance can degrade very quickly - often exponentially - with
> the workload.
>
> With EVA, performance will degrade for pretty much every system who's LUNs
> came from the same storage group. But the upside is that the performance
> degrades much more gradually, and much more linearly with the workload (due
> to the statistics involved in the larger number of spindles).
>
> If performance begins getting bad, adding 10 (or more) disks to the group
> is fairly simple, and soon you've alleviated all the I/O bottlenecks. And
> it's all transparent (operationally speaking) to the host servers that are
> using the LUNs. Performance just gets better.
>
> That beats the hell out of manually moving files, or partitioning
> databases/tablespaces among lots of different LUNs to balance I/O.
>
> I guess if I understood why someone felt the really needed a VTDPY-like
> tool (i.e., what problems do they need to solve with it), it might help.
> But to manage performance I've almost never actually needed that tool.
-- Jilly - Working from Home in the Chemung River Valley - Waverly, NY - jilly@clarityconnect.com - Brett Bodine fan - Mark.Jilson@hp.com - since 1975 or so - http://www.jilly.baka.com
- Next message: Tom Linden: "RE: Why repost this? (was Re: [OT] Humour)"
- Previous message: fernando.vallarino_at_oca.com.uy: "Re: Problems with SYS$GETRMI"
- In reply to: jlsue: "Re: EVA-5000 performance monitoring"
- Next in thread: Hal Kuff: "Re: EVA-5000 performance monitoring"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Relevant Pages
|
