AlphaServer ES47 7/1000 - Need a SUPER fast (big) io device!

[RBrown_1]

Sorry about that, Microsoft is helping me even less than I thought it was helping you.

My first attempt to view your attachment showed me a picture of a paperclip!

[Edmundo T Rodriguez]

...

0 for the comment + 1 point for the joke

...

[Robert Gezelter]

Edmundo,

Thank you for the attachment. It clarifies some things and leaves a great many questions unanswered.

1TB in an hour translates to 284KB/sec (1,024MB/(60*60). Converted to bps, that translates to 2.27Mbps. Granted this is a sustained rate, which must be discounted significantly, but offhand, I am not sure I would be so fast to attribute the problem to the SAN.

I would recommend a deep examination of that backup procedure. I have seen more than my share of backup and other end-of-day procedures that were effectively fratricidal in one way or another.

If this is the case, adding a disk drive may very well make the problem worse, not better. I would recommend a careful, in-depth review of the backup/end-of-day procedures, either in-house or out-of house (Disclosure: We provide services in this area, as do others who regularly contribute to this forum).

As a side note, I would also be interested in seeing the CPU utilization data corresponding to the same time period as the IO queue length chart.

There is a good chance that working the problem without looking at the script in detail is a futile exercise.

- Bob Gezelter, http://www.rlgsc.com

[Andy Bustamante]

As Bob (Robert G) points out, it's time to collect and review system metrics. Before deciding on any approach, identify what the system is really doing, what bottlenecks may exist and potential solutions.

Look at CPU load, processes in the COM state, memory used including XFC numbers.

Some potential options including adding a third set of shadow set members or clustering another server to manage the backup operation while the production server continues to service user and report loads. Before deciding an an approach, identify the current system metrics.

[Edmundo T Rodriguez]

Trust me Robert (and all others), you donâ t want to see (less analyze) a 3rd party procedure of 3334 lines, where you can see all kind of routines dealing not only with OpenVMS DCL commands but also MUMP, SQL.

Sorry to say like this, but there is nothing to be improved here.

This is a fault proven procedure running here and others places for 7 years.

Attached you will find more graphics snap-shots of the AlphaServer ES47 for the same period.

You will notice (if you can rationale: performance vs. load) that the only things bugging a little bit (during only the high picks of day) is physical memory, which is not the problem during the backup time from 8:00 PM to 1:15 AM

[Edmundo T Rodriguez]

CPU vs IProcesses

[Edmundo T Rodriguez]

CPU Process Types

[Edmundo T Rodriguez]

Page Faults Rate vs IProcesses

[Edmundo T Rodriguez]

Page Faults Rate vs IProcesses

[Jon Pinkley]

>>1TB in an hour translates to 284KB/sec (1,024MB/(60*60).

This was a mentagraphical error off by a factor of 1024.

1TB in an hour is sustained 291.3MB/sec. (1024*1024)MB/(60*60)sec ( if TB = 2^40 bytes)

So a 2MB SAN can be a limiting factor if 1TB/hour is your goal.

A single spindle mechanical PC drive won't cut it, even if you could connect it to your ES47.

Edmundo,

You said you are using shadowing, but it isn't clear that you are using it for backups.

If you are backing up 1TB in 5.5 hours, that is about 53MB/sec, which is reasonably good.

Why is 5.5 hours too slow? If you are using volume shadowing with split members for backup, that is essentially using your storage a snapshot. If you are going to use shadowning, make sure you have the HBVS patch that supports mini-copy on 7.3-2, and make sure you ensure that minicopy gets used when you reintroduce your shadowset members.

Jon