Question about disk performance.

lcsw2013

Hi all,

Could anyone out there tell me how you have your environment setup to provide best possible performance?

1. We are all virtual.

2. We are on SAN storage.

3. We are a decently large environment with a pretty large database. and several additional pollers.

We believe the problem to be related to SAN or Disk configuration. But as you might of guess storage and windows teams say otherwise. What is the best way to setup virtual disks that reside on a SAN for optimal performance? Do they need to be dedicated for the database? Do they need to be tier 0 or higher on tiered storage? How can they be optimized for speed and not storage? And what numbers should we be looking at?

Anyone that can help with parameters and suggestions would be greatly appreciated.

thanks,

leandro

Find more posts tagged with

speed

website

Disk

Performance

sql

npm

Storage

san

Accepted answers

lcsw2013

So I found the problem. Solarwinds is picky as hell when it comes to working with VM machine that's hooked to a busy SAN. Taking a slice of the SAN's avaliable SSD storage, And placing our VM in the FAST policy's of the SAN fixed the issues. Counters that where in the seconds dropped to a max of a 150 ms and that's only when we really push it. Otherwise it stay's below 100ms in latency. IOPS where brought up and overall the database (While not optimal) is working within tolerance levels. IPAM and SAM where causing too many requests. Optimizations and customizing took care of this.

If there is one thing we learned it is that when you have a VM, SAN cannot police the traffic and disk usage. This will come at performance cost to the application. On a SAN, you must still be dedicated fast drives and your storage team needs to have your VM's in whatever fastest storage policy they have configured.

After storage made the move we have been performance problem free for over 2 weeks and counting. And everything completes way quicker. This is now resolved.

All comments

lcsw2013

So we found from storage team that we where on a large pool of drives that where dynamic tearing. In other words low performance files go to regular sata drives while higher demand files go to 15k sas drives. A small slice of our LUN is SSD's which isn't enough to really provide much relief.

Storage team switched out lun so a much bigger portion is flash storage (SSD's) With the remaining portion tier 0 fast spindles. The updated Lun policy to fast policy (Not sure exactly what this means). But this alone seems to have cut in half all metrics and performance has improved decently.

Wait times that where in the seconds is not down to mas of 150ms. Not optimal but much much faster response. And given that this is SAN fiber connected storage this falls within expectation. Windows team did detect few errors recorded on the fiber channel being used and moved us to a different channel as well.

Next stage is that we have a 250gig database with a sql that only has 70gigs of memory allocation. Meaning that page life expectancy and cache usage is high, but overflowing to paging operations. Fixing disk performance fixed the paging delays but did nothing for the cache and ram buffer issues. We have a change in place to add more ram.

This last item goes to any SQL experts out there. How can we improve efficiency of data flow and how sql handles data within the database? I find that alot of times it's working in blocks. An operation is requested. SQL takes a chunk of that operation works on it delays the rest and continues working on it instead of a real time take the whole request work it and send it back once completed. Seems to me like there is room for efficiency improvement and I need help here.

lcsw2013

After storage made the move we have been performance problem free for over 2 weeks and counting. And everything completes way quicker. This is now resolved.