This post is more than 5 years old
4 Posts
0
13605
Storage Pools 3 Hybrid Arrays
I have 2 6500ES's and 1 6100XS being utilized by 8 ESX hosts running about 130 VM's. Performance is important, but I just migrated my MD3000i data onto the 6500's, so I'm already seeing an improvement...the SSD's are awesome. :)
I have 3 questions
1.) Should I make 1 Pool or 2? I currently have them in 2 pools, however, I am wondering if just making 1 pool and letting the EQL load balancers do all of the guess work for me... I don't want to have to check IO usage between pools like I have been and manually or using storage DRS if I don't have to. IO based storage DRS requires turning on storage IO control, which is not best practice according to the EqualLogic VMWare Best Practice Documentation - I'm guessing because it could hinder the SAN's load balancer decisions.
I believe I read that best practice says 2 pools, but I don't know if it takes 3 hybrids into consideration.
With the exception of the VM's our data is pretty transient. We receive a small amount of data, create a large amount of data, duplicate it, back it up, deliver it, then it sits for a period of time before getting archived & purged.
2.) I am seeing higher than expected read latency on the 6500's when IOPs drop below 500/sec. it's not crazy high, but hovers around the 8ms range. I turned off delayed ACK which brought the latency down from 12-15, and want to turn off TCP offloading next maintenance window. But, is this normal? My 6100 usually stays below 2ms read and write
3.) Firmware updates - is it safe to do while there is activity? Or do I really need to shutdown all of my ESX hosts before doing it? The documentation says to do it when there is low utilization... Does that mean that status in SAN HQ says low? or what is considered low? Does the same apply for the hard drive firmware updates?
Thanks in advance!
David
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
1
January 15th, 2014 08:00
Re: 1. I would suggest you keep them at two pools. Only because of the size difference between the 6500's and 6100s. You could go to 3 pools, but they you do run the risk of having to move VMs between them to balance IO.
Re:2. I would also try turning off LRO, Large Recieve Offload.
Re: 3. Upgrades. Low I/O means when the array isn't being heavily used. So yes, a low I/O reading from SANHQ would show that. If your business isn't active 24x7 then it's the normal after hours weekend time frame. Most important is setting the login_timeout to 60 on ESXi servers, and setting the disktimeout value to 60 in the VMs. With recent firmware the time it takes to failover and start responding is usually under 30 seconds. But not always, an can be nearer to 60 seconds under load. Also critical is to make sure the switch ports come up immediately on transistion. (AKA portfast) This way when the CM fails over, the ports come online right away. Also, being able to handle a FW upgrade also means you can handle a CM failover on a HW problem.
Disk FW upgrades. Yes. The upgrades are done one drive at a time. Then the drive is power cycled, to update the FW. That time is very quick, and only I/O to that drive is paused, then resumed once the drive comes back online. But again, best time is low I/O.
In a perfect scenario, yes doing upgrades in a Maint window would be ideal. At least shutting down the VMs. In that off chance something unexpected happened.
Since this is ESXi are you running at least 6.0.7 firmware?
Regards,
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
January 15th, 2014 15:00
Re: 1. Spindle speed, nor drive type, SSD vs. SATA vs SAS are NOT taken into account. By default, the space will be distributed by capacity of the members in the pool. So the 6500's will have a much larger share of the pages and therefore do more of the work. When the sustained latency of a member is signficantly greater than another, that's when hot/cold page swap occurs to bring the relative latency into balance between the members. . So I would stay with two pools in your case.
Re: MEM. Depending on the version of MEM, yes it does.
re: VMware tools. It's supposed to with Windows clients, but I always suggest verifying it. It does not for other OS's like Linux or Solaris, etc.. Those would have to be done manually
XDI_David
4 Posts
0
January 15th, 2014 15:00
Follow-up question (clearer way to ask #1) - are the spindle speeds taken into account for the performance load balancer? - so in the case of 2 slower hybrids and 1 faser hybrid... will the most active data eventually go onto the faster hybrid? or will they stay on the original hybrid since the SSD's will be able to keep up with demand.
XDI_David
4 Posts
0
January 15th, 2014 15:00
Thanks for your response.
#1 - If I go with 1 pool, and after the initial space load balancing occurs, will hot/warm blocks/slices move from the 6500 to the 6100 or will they just move from spindles to ssd, and back?
#3 - does the MEM install auto-update the timeout to 60 seconds? and doesn't the VMWare tools increase the disktimeout on the VM's?
I am currently running 6.0.1 on the 6100 and 6.0.2 on the 6500's - thinking about upgrading to 7.0.1, unless you'd recommend that I wait.
Thanks in advance for your followup.
David
XDI_David
4 Posts
0
January 17th, 2014 16:00
I just wanted to post an update for the benefit of others who may have the same or similar issue with higher than expected read latency.
Turning off delayed ack reduced my read latency from ~15ms to ~8ms
Turning off LRO (Large Receive Offload) reduced my read latency from ~8ms to ~2ms
Now I just need to figure out why I can't get my Queue Depth to go below 1 on the 6500... Any thoughts Don? SANHQ is reporting almost a flat line at 1. I have only seen it drop below 1 for a VERY short period of time (1 poll cycle) a few times a day