Intel DC S3700 Review – a full Performance Characterisation

Myce/OakGate Reads and Writes Tests

Here is the specification of the tests –

The tests are designed to show the Random and Sequential,
Read and Write, performance metrics for different combinations of Queue Depth
and IO size.

Here are the results –

Random Reads

You can see here that Read IOPS does not scale beyond a
queue depth of 16/32. You can also see that the level of IOPS for 4KB writes,
at a Queue Depth of 32 is slightly over Intel’s specified value of up to
75,000.

 

You can see here that bandwidth doesn’t scale beyond a queue
depth of 16/32.

You can see here that Read Latency does not change over a
queue depth of 32. I think we can conclude that the Intel effectively operates
at a maximum queue depth of 32.


Random Writes

You can see here that there is no effective scaling for
Random Writes IOPS above a queue depth of 2.  You can also see that the level
of IOPS for 4KB Writes, at a Queue Depth of 32 is below Intel’s specified value
of up to 36,000.

Similarly you can see here that there is no effective
scaling in Random Writes bandwidth above a queue depth of 2.


Sequential Reads

You can see here that, just as for Random Reads, there is no
effective scaling in Sequential Read IOPS beyond a queue depth of 16/32.

You can see here that there is no effective scaling in
Sequential Read bandwidth beyond a queue depth of 16/32.

You can see here that there is no change in Sequential Read
latency beyond a queue depth of 32.


Sequential Writes

You can see here that there is no effective scaling for
Sequential Writes IOPS above a queue depth of 8.

You can see here that there is no effective scaling for
Sequential Writes bandwidth above a queue depth of 8.

 

You can see here that there is no change in Sequential
Writes latency beyond a queue depth of 32.

Remember back in the SNIA Throughput Test I said we’d take a
look at Intel’s specified Throughput numbers of –

‘Sustained Sequential Read: Up to 500MB/s’ and ‘Sustained
Sequential Write: Up to 460MB/s’

Let’s have a look now.  To do this I have run a simple test:
i) Purge the Device ii) For 2 hours perform sequential 1024K (block size)
writes at a Queue Depth of 32 iii) For 2 hours perform sequential 1024K reads
at a Queue Depth of 32, and in this test I have enabled Write Caching.  Here
are the results:

You can see the write speed falls slightly after
approximately 700 seconds as the drive reaches a Steady State.  Let’s zoom in
on part of the graph line, which is in Steady state to take a closer look.

You can see a cyclic pattern again and the bandwidth spends
most of the time at around 390MBs but then falls to around 355 MBs for the rest
of the time.  This doesn’t compare well to Intel’s spec. of – ‘Sustained
Sequential Write: Up to 460MB/s’, but I notice that an explanatory note is
tagged to the Bandwidth Performance values, which says – ‘Performance Values
vary by capacity and form factor’; so I assume that the up to 460MBs applies to
the 400GB and/or 800GB models.   

Here is the graph for Read Bandwidth performance. You can
see that the specified – ‘Sustained Sequential Read: Up to 500MB/s’ is being
exceeded.

For interest, here is a zoomed in picture of part of the
graph line.


Now let’s head to the next page, to look at the results
for the Myce/Oakgate 4K Mixed Reads/Writes Tests…..

 

NEW PAGE NEW PAGE NEW PAGE NEW PAGE