Six port network card

 When four ports isn't enough, use six?  I have seen dozens of quad port NICs but never a six port.

Silicom Pe2g6i35 use the Intel's i350 chipset.

https://www.silicom-usa.com/pr/server-adapters/networking-adapters/gigabit-ethernet-networking-server-adapters/pe2g6i35-server-adapter/

Enterprise SSD tear down

I had some failed drives that I was unable to warranty, so off to recycling they go!  I took a few pictures as I thought they were interesting; there is so much more going on in these than a standard desktop drive.

Notice the super caps!

 

RAID performance comparison

 Time to look at an age old question....when one only has a few drives, what RAID configuration is best for performance. Conventional wisdom is:

RAID0 is the fastest as two drives are load sharing.  Also known as "striping," the data across two drives, in theory two drives are splitting the load equally.

RAID1 is next fastest.  Also known as "mirroring:", all data is being written to both drives but either or both data can provide reads.  

Finally RAID 5 being the slowest.  In a three drive system the data is split in half, two drives get the data chunks and a third drive gets a calulated parity bit that can be used to reconstruct either of the data chunks.  Those three pieces are rotated among the drives in the member set.  The calulation of the data is what takes time, in theory more time than the previously mentioned methods.

First test: HP Proliant ML350 G6, SmartArray P410 with 1gb battery backed cache.  The hard drives are HGST 7200rpm SATA drives.  This machine is running VMware ESXi, the RAID volume was created and a virtual hard drive was created on the datastore for that RAID volume using thick provisioning.  The Atto test was run twice just to make sure the results were consistant.

 

Results:  In this test RAID5 out performs both RAID1 and RAID0 in both reads and writes.  RAID1 had better read performance than RAID0.  However RAID0 was but was faster than RAID1 in writes.

Second Test:  Dell PowerEdge R620, Perc Mini H310.  The hard drives are Intel Enterprise 200gb 3Gbps SSD's.

RAID0 (stripe)

RAID1 (mirror)

RAID5 (span with parity)...holly cow!  Do not do Raid5 w/ a Perc H310!

RAID5(span with parity)(RAID Controller w/ cache)

Results: RAID0 was much faster writes, but a tiny bit slower in reads when compared to RAID1.  Also in this test RAID5 was both faster than both RAID 0 and RAID1 in both reads & writes.  However it should be mentioned that the machine used for the 2nd R5 test was slightly different, it had a slower CPU (e5-2640 vs e5-2650) it also had a better RAID card (a Perc H710 vs H310).  The two RAID cards both have the LSI 2008 chipset, but I the H710 has 512mb cache and the H310 has 0mb.

Third test:  not apples to apples, but interesting.  The speeds of test number two were a bit low, so I tried a Samsung Evo 840 (CIRCA 2014).  The first test is on a Windows 10 PC, the 2nd test in the same Dell R620 as above, but in a RAID0 configuration.

Results: Early SSD's leave much to be desired in terms of performance.  The Evo 840's are TLC drives.

So it would seem that, atleast from these tests.....a three drive RAID5 out performs a two drive RAID0 or RAID1 configuration.