Primary Storage

High Capacity Flash on the All Flash DS8880

By Nick Clayton posted Fri January 05, 2018 02:26 PM

  
In September 2017 we introduced a new class of Flash for the DS8880 with a new 3.8TB High Capacity Flash Drive. These drives offer a lower cost point than the High Performance Flash Drives and are optimised for capacity. The read performance of these drives is similar to the High Performance Flash Drives but the maximum write throughput is lower.

With the hardware RAID in the DS8000 the High Capacity Flash Drives still give very good performance. High Capacity Flash can handle many typical production workloads, but may not have sufficient sequential throughput capability for an intense batch type workload. If the workload access density is such that the 3.8TB drives can handle the peak workload, then using these drives by themselves will give the best All Flash configuration.

The list price of the High Capacity Flash drives by themselves is around 50% of the cost of the High Performance Flash drives. However, to understand the cost of using these in an Enterprise storage system like the DS8000 we need to consider the overall system costs including infrastructure and advanced function. If we take a sample 200TB configuration, then we can see that the overall system cost using High Capacity Flash is around 66% of the cost of a system using the 3.2TB High Performance Flash, which would previously have been the best choice from a cost perspective.

Single Tier Flash Cost Comparison

Easy Tier has also been enhanced to enable the High Capacity and High Performance Flash drives to be combined in the same storage pool by dynamically mapping different drive types to the three tiers in Easy Tier. Easy Tier will treat the High Performance drives as a higher tier than the High Capacity drives. This will concentrate the workload on the High Performance drives in the same way as for Flash and Enterprise drives. The Easy Tier algorithms have been changed to promote both intense sequential activity as well as random workload to the High Performance Flash.

If the workload access density is higher than the 3.8TB drives can handle and the workload is skewed then with the new Easy Tier function, a combination of the 3.8TB drives and a percentage of smaller (e.g. 400/800GB) High Performance Flash drives can be very attractive. In general it will not make sense to combine the larger 3.2TB High Performance drives and the 3.8TB High Capacity drives, as these are very similar in size and hence can support a similar access density of workload.

The first step to deciding a sensible configuration with a combination of High Performance and High Capacity Flash drives is to decide what the largest workable drive would be for a single tier High Performance Flash configuration. Disk Magic should be used to calculate this. One the single tier High Performance Drive size is known, then you can look at possible multiple tier configurations using a combination of some percentage of smaller High Performance Flash drives with the rest of the capacity using the High Capacity Flash drives.

Exactly what percentage and size of drives would make sense will depend on the skew of the workload. It will depend on the MB/s skew rather than the IOPS as this is typically the limit for Flash drives. Easy Tier reports the skew information for workloads already on the DS8000 as shown in the sample skew report below. If we were to take the top 7.5% of the capacity then this would have 50% of the MB/s and around 90% of the random reads and the random and sequential writes.

Skew Curve Example

If this workload could use 1.6TB Flash drives as a single tier as determined using Disk Magic, then a 207TB capacity (usable RAID6) single frame DS8886 configuration might look like the one below.

Capacity Magic Display of Single Tier Flash Configuration

Using a mix of 400GB High Performance Flash Drives and 3.8TB High Capacity Flash Drives we can create a configuration as shown below. This has 222TB capacity (usable RAID6) and 7.5% of the capacity as High Performance Flash. This has a very similar overall throughput, and with the skew given above would be a sensible alternative configuration for the same workload.

Capacity Magic Display of Multiple Tier Flash Configuration

This sample two tier configuration using the High Capacity Flash configuration is 29% lower list price than the single tier configuration using High Performance Flash. Looking purely at the drives and excluding the DS8000 servers, host adapters, infrastructure and microcode features etc the cost is 40% lower.



This example shows that the combination of smaller High Performance Flash drives with High Capacity Flash drives managed by Easy Tier can provide significant cost benefits compared to using High Performance Flash drives by themselves. The key points to remember are:

  • Use the 3.8TB High Capacity Flash drives as a single tier if they can handle the workload.

  • If combining High Performance and High Capacity Flash then use a smaller High Performance Flash drive than you would if used by itself.

  • Disk Magic with the Easy Tier Skew reports from STAT or the R8.3 GUI can help model performance for a known workload.




#EnterpriseStorage
#DS8880
#Flash
#DS8000
#Flashstorage
3 comments
4 views

Permalink

Comments

Tue January 09, 2018 07:07 AM

Hi Nick,
Thanks for your quick feedback.
I appreciated to be independent from the GUI. It is also straight forward when utilizing the DSCLI to get the data directly on my laptop.
Nonetheless it is now as it is.
Thanks and cheers
Werner

Mon January 08, 2018 02:25 PM

Hi Werner,

With R8.3 we have taken the valuable information from the standalone STAT tool and integrated this into the DS8000 GUI reports. The requirement to install another piece of client software to do the analysis was not universally appreciated to say the least and so we took the step to remove this.

It is also possible to download a set of CSV file reports from the GUI if you would like to do some offline analysis of the Easy Tier behaviour on one of your clients machines. These are the same CSV reports that were produced by the STAT tool and aim to provide everything you might need to understand the Easy Tier behaviour.

Sun January 07, 2018 10:48 AM

I regret that the STAT tool is no longer a CLI based stand-alone interface tool to analyze Easy Tiering data.
Nick‘s documents and hints and tips are always interesting and helpful.