Calculating ESS Array Endurance

The ESS Raid Designer includes logic to calculate what to expect in terms of SSD life. This document shows how these calculations work.

Examples:

We will calculate values for both 25% compressible blocks and 60% compressible blocks. 25% is typical for general purpose file systems. 60% is typical for text databases. This calculation scales to any number of drives and any raid levels. The AWPD remains contstant regardless of drive size. The TBW grows with the number of data drives.

We can calculate wear for a single drive because ESS does parity writes on "perfect stripes". Each parity drive gets exactly the same number of writes as each data drive. Thus, RAID-5 and RAID-6 have the same wear as RAID-0, just less capacity.

Step 1: Calculate the "Total Writes Available" of a Single Drive.

         3000                      Typical Endurance for TLC Flash
x 1.26   3780                      1 DWPD SSDs have 26% Over Provisioning

Step 2: Calculate Raw Block Fill Levels

     25% Fill           60% Fill
         0.95               0.95   This is the raw reserved fill space.
x 0.85 = 0.81      x 0.85 = 0.81   The model is built for 85% full file systems.
x 0.75 = 0.61      x 0.40 = 0.32   Multiply by the compresibility of the blocks.
         1.53:1             1.17:1 This is the raw Write Amp for ESS.
x 0.75 = 1.14:1    x 0.40 = 0.47:1 This is our final Effective Write Amp.

Step 3: Calculate TBW.

         3780               3780   From Step 2, the SSD available writes.
/ 1.14   3315      / 0.47   8042   This is the TBW for a single data SSD

Step 4: Calculate AWPD.

/ .95    3489      / 0.95   8466
/ 365    9.56              23.20    Years of expected life
/ 5      1.91               4.64    Array Writes per Day (AWPD)

Discussion

Typical datacenter 1DWPD SSDs have a Write Amp of about 2.1:1. This, with 26% OP yields 1 DWPD.

When you put this into an array, you then incur the overhead of parity writes. Each write to RAID-5 becomes two writes. Each write to RAID-6 becomes three writes.

The random pattern also causes the SSD to slow down as the FTL has to start doing garbage collection.

ESS off-loads the FTL to a layer in front of the RAID logic. This eliminates the parity "multiplier" for writes. Compression then increases free space, which lowers the block-level write amp. Then compression lower the writes again.

The result is that, for many workloads, ESS can extend SSD life 5X or more.