HighPoint RocketRaid 4520 Raid Card SSD Benchmarks and Review

Posted on Feb 21, 2013 by Paul White

I took a gamble and decided to try Highpoint's latest flagship Raid Card ( RocketRaid 4520 ) for my new web server.  Naturally when I get a new piece of hardware I like to put it through some benchmarks to see if it really performs the way it should.  When this combined with your Operating System's Buffering and Caching this card is Excellent, but when Direct I/O is tested is falls short.  To make a long story short, the Rocketraid 4520 does not perform well for IOPS at 4K.  I have already contacted HighPoint and they inferred that their engineering team is aware of this, and are working on a new firmware that should fix the issues.

Highpoint RocketRaid 4520 Overview

The RocketRaid 4520 is a beefy card boasting 6 Gbit SAS/SATA connections.  2 internal Mini-SAS ports ( SSF-8087) provide connections for up to 8 devices without the use of SAS expansion cards.  It has a dedicated RAID-on-Chip processor and 512MB of DDR 3 cache memory.  Its a low profile PCI Express 2.0 x8 card allowing it to fit in just about any configuration.  It supports RAID 0, 1, 5, 6, 10, 50, and JBOD.  It also supports drives larger than 2TB, and is compatible with Windows server 2003, 2007, 2008, 2012, and Windows Vista, 7, 8, as well as major linux distributions and FreeBSD.

What comes in the box?

Highpoint includes the Mini SAS cables, and both a low profile and regular mounting bracket.  Manual, Drivers disc and other papers are also included.  I also bought the BBU separetely, and it comes with its own low profile mounting bracket, and cable to attach it to the RAID card.

Installation

Installation was straight forward.  Though I decided to use the cables already installed in my server as the ones provided by Highpoint are a bit bulky.  When you are installing anything in a 1U rack server, space is not always abundant.

HighPoint RocketRAID 4520 Boot Manager

One thing I like about this card is the boot manger to setup your RAID arrays is very intuiative.  Below are some screen shots of what you will see when setting up an array.

RocketRAid 4520 vs Other Raid Cards

 When comparing it against other cards it appears to be the best value

Card	HighPoint RocketRaid 4520	Adaptec 6805	LSI MegaRAID 9261-8i
Interface	PCI Express 2.0 x8	PCI Express 2.0 x8	PCI Express 2.0 x8
Devices	8	8	8
Port Speed	6 Gbit	6 Gbit	6 Gbit
Processor	Marvell 88RC9580	PMC-Sierra PM8013 Dual Core RAID	LSISAS2108 RAID-on-Chip
Cache	512MB DDR3 ECC Protection	512MB DDR2 (667 Mhz)	512MB DDR2 (800 Mhz)
Raid Levels	0, 1, 5, 6 10, 50, JBOD	0, 1, 1E, 5, 5EE, 6 10, 50, 60, JBOD	0, 1, 5, 6 10, 50, 60
Low Profile	Yes	Yes	Yes
Backup Power	Optional Battery HPTBBU-05	Optional Capacitor AFM-600	Optional Battery LSIiBBU07
Price Point	$429.99	$549.99	$484.99

Benchmarking the RocketRaid 4520

I bought this card for use in a multipurpose Web Server ( IIS 7.5, MySQL, Email, Stats ).  My Final Production setup will be 6 x 250GB Samsung 840 SSD drives, and 2 x 2TB WD Green SATA drives.  The SSDs will be configured into an array, and the SATA drives will be used for daily backups.  But for our testing I have installed the OS ( Windows Server 2012 Essentials ) onto a 160 GB SATA drive.  Then I will configure the 6 SSD drives into various RAID levels and stripe sizes testing each one to see what kind of performance I get.  The Software used to Benchmark these will be ATTO 2.47.  My goal is to determine what the optimal RAID setup would be with these SSD drives, or if I should sell the card and upgrade to something faster.  

Benchmarking Platform

• Supermicro 1026T-UF
• 8 x 2.5" Hot swap Trays
• Dual Xeon 5570s( 2.93 Ghz Quadcore with hyperthreading )
• 48 GB of DDR3 Memory
• 560 Watt Power Supply
• 6 x 250GB Samsung 840 SSD drives
• 1 x 160GB SATA HD ( primary with OS )
• Windows Server 2012 Essentials

Why the Samsung 840 SSD and not 840 Pro?

The only reason to get the 840 Pro would be for fast Writes Especially IOPs at 4K, or because you are worried about wearing out the flash memory since the 840 is TLC based and the 840 Pro is MLC based.  Theoretically the 840 Pro can handle 3 x the writes of the 840.  But in my case I have already calcuated what kind of wear I will be putting on these drives and wearing out the drives is not likely.  A single drive will last 11 years in my server, and considering I will be using these in a stripped array, this multiplies the expected life by the number of drives in the array.

About ATTO Disk Benchmark

We will be using ATTO 2.47.  We are using the default settings which are Transfer size 0.5KB - 8192.0 KB, Total Length 256 MB, Overlapped I/O and Direct I/O enabled. Queue Depth of 4.  These are the typically settings most websites use when benchmarking the drives, but they don't tell the whole story.  

1 setting in particular ( Direct I/O ) deserves more attention.  When Direct I/O is checked it will bypass the Operating system's caching and buffering.  Each version of Windows has made imrpovements in it how it cache's and buffers frequently accessed data.  When Direct I/O is unchecked, the benchmark will reflect any performance benefits the operating system brings to the drive.

Another notable setting is the Queue Depth. This is the number of files to be Queued at once.  The higher the value the more files that can be transferred at once.

Our first benchmarks will leave Direct I/O enabled to get the drive's stand alone performance.  This also helps to standardize our performance against other benchmarks around the web.

HighPoint RocketRaid 4520 Benchmarks

Using ATTO I benchmarked the RAID card using various raid levels and stripe sizes.  I also ran some benchmarks that were direct IO ( which bypass any buffering or caching by the operating system ), this should show the cards true stand alone performance.  Benchmarks were performed with 250GB Samsung 840 SSD drives ( Not the PRO versions ).  

1 x Samsung 840 SSD Benchmark on RocketRaid 4520

2 x Samsung 840 SSD RAID 0 with 64KB Stripe size Benchmark on RocketRaid 4520

4 x Samsung 840 SSD RAID 0 with 64KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 0 with 64KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 0 with 16KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 0 with 1024KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 10 with 64KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 5 with 64KB Stripe size Benchmark on RocketRaid 4520

6 x Samsung 840 SSD RAID 6 with 64KB Stripe size Benchmark on RocketRaid 4520

RocketRaid 4520 Benchmark Chart on Samsung 840 SSD drives

Number of SSDs	Raid Level	Stripe Size	ATTO Direct IO	IOPs Read @ 4KB	IOPs Write @ 4KB	Max Read	Max Write
1	JBOD	NA	enabled	20,362	17,457	537 MB/s	295 MB/s
1	JBOD	NA	disabled	230,647	182,876	3,688 MB/s	3,689 MB/s
2	RAID 0	64KB	enabled	20,060	18,476	1,060 MB/s	545 MB/s
2	RAID 0	64KB	disabled	251,286	205,934	4,252 MB/s	3,286 MB/s
4	RAID 0	64KB	enabled	20,361	18,986	1,993 MB/s	1,058 MB/s
4	RAID 0	64KB	disabled	266,863	198,373	4,273 MB/s	3,910 MB/s
6	RAID 0	64KB	enabled	20,769	19,750	1,723 MB/s	1,556 MB/s
6	RAID 0	64KB	disabled	205,422	180,572	4,273 MB/s	3,872 MB/s
6	RAID 0	16KB	enabled	19,750	19,081	1,147 MB/s	1,322 MB/s
6	RAID 0	16KB	disabled	232,115	181,340	4,284 MB/s	3,702 MB/s
6	RAID 0	1024KB	enabled	17,328	18,313	2,214 MB/s	1,532 MB/s
6	RAID 0	1024KB	disabled	218,359	181,398	4,284 MB/s	3,739 MB/s
6	RAID 10	64KB	enabled	20,514	26,505	1,669 MB/s	785 MB/s
6	RAID 5	64KB	enabled	41,642	33,333	1,750 MB/s	963 MB/s
6	RAID 6	64KB	enabled	19,750	15,538	1,484 MB/s	813 MB/s

Final Thoughts on Benchmarks

When ATTO is not forced into Direct I/O mode benchmarks are great,  but it when the Operating System's caching and buffering is taken ouf the picture the performance plumets.  Hopefully HighPoint releases a Firmware update to fix the poor IOPs @ 4KB.  The Samsung 840 SSD are cable of 96,000 IOPs read.  Granted I don't expect to hit 500,000 IOPs with a 6 drive RAID array, but I do expect to hit at least 250K.  However considering that the Operating system will be implimenting its own caching and buffering, I feel that my numbers are likely to be closer to the Direct IO disabled values.

The performance of this card by no means on the bleeding edge.  At this point I think the LSI Cards with their Fast Path addon Module is probably the best way to go for those wanting serious performance out of their SSD arrays.  But I have to remember I am building a multipurpose web server, and not a monster gaming rig.  

Discussion

name
Email	Needed to confirm comment, but not made public.
Website

	Email me when someone comments on this blog

When you Post your Comment, you'll be sent a confirmation link. Once you click this link your thoughts will be made public.. Posts that are considered spam will be deleted, Please keep your thoughts and links relavent to this Article