Equallogic Load Balancing

October 17, 2013 Jordansphere Equallogic

A have written a very brief overview of the techniques Equallogic use for load balancing in a storage pool. These can be split into 3 different components:

1) NLB – Network Load Balancer

Depending on network load, Equallogic can redistribute ISCSI connections across Ethernet ports. This is accompolished automatically by using an inbuilt feature of iSCSI called redirection. The iSCSI target alerts the iSCSI initiator to log out and close a connection, then immediately start another session via another IP address. 

Hosts connect to the Equallogic group via an iSCSI group IP address so all of this is done automatically. This feature is essential for such tasks as moving members in and out groups or moving pools.
From the screenshot below you'll see some informational logs displaying the closing and opening of iSCSI connections.  In this task I am actually moving a member into a pool and NLB is doing the rest automatically.

 


2) CLB – Capacity Load Balancer

The CLB ensures the capacity over all members in a pool is evened out appropriately. This is measured on a percentage basis of each members capacity and ensures one member isn't completely overloaded. It also keeps relevant free space for potential maintenance and replication.

 

3) APLB – Automatic Performance Load Balancer

I believe this was introduced in firmware version 5.1.x and is on by default. Essentially Equallogic optimizes the utilization of a SAN performance pool by using iSCSI latency as its metric.

The APLB is intelligent enough to react to workloads in realtime and remedy any imbalance by moving the I/O to another member with a lower latency. This task runs every few minutes. If there are more than 1 potential target member then Equallogic will choose its new target by using factors such as disk speed, no of disks, RAID  etc

Data is grouped into "hot" and "cold" spots and distributed over the members accordingly. Take my Outlook inbox as an example. Now, I have been using my inbox for about 5 years and its around 4GB in size. The majority of the data I havent even looked at in years so has hardly been accessed. This data would classified as "cold" as it is never (or rarely) accessed.

 

 

Can you mix members with different RAID/disk speed/Network links?

I have taken this directly from Equallogic documentation (as of Oct 2013). This should answer any questions:


The following best practices should be considered for storage pools unless the automatic performance load balancer (APLB) is disabled: 
• Mixing arrays of different drive speeds and types is fully supported and is the primary reason APLB was created. 

• Do not mix arrays with different controller speeds (1GbE, 10GbE) within a single pool unless they are each running unique RAID policies. 

• To override the automated performance decisions for a specific volume, indicate a “preferred” RAID type for that volume. If that RAID type exists within the pool, the volume will reside on those arrays that match the preferred RAID policy for the volume. 

 

If the APLB is disabled, the following recommendations apply: 
• Do not mix arrays with different drive speeds within a single pool unless they are running a unique RAID policy 
June 2013 Dell EqualLogic Configuration Guide v14.2 5-38 

• Do not mix arrays with different drive technologies (SATA, SAS, SSD) within a single pool unless they are running a unique RAID policy. 

• Do not mix arrays with different controller speeds (1GbE, 10GbE) within a single pool unless they are each running unique RAID policies

 

 

If you want a more indepth analysis of the Equallogic LBs then please download their official documentation.

NLB APLB CLB equallogic load balancing,

Powered by WordPress. Designed by elogi.