The Rocket RAID 2640X4 is Highpoint’s PCI Express SAS RAID controller card with four internal ports. Model number 2642 offers two internal and two external ports, while model 2644 was designed with four external ports; there is also a Rocket RAID 2680 card with eight internal ports. All of these products are based on four-lane PCI Express interfaces, but they may also be operated on x1 PCI Express connections at reduced performance. The Rocket RAID 2640X4, which we received for review, is a low profile card that is based on a Marvell controller and four individual SFF-8482ports. The most interesting detail about this storage controller is its price tag of less than $150. It even comes with a three-year warranty.
Highpoint supports the common RAID levels 0, 1, 10 and 5, native command queuing, multiple arrays per controller, array roaming for easy transfer to other Highpoint RAID controllers, online capacity expansion, RAID-level migration on the fly, email notification, and I2C-based enclosure management. The feature list is as long as the lineup of supported operating systems, including Windows 2000 and up, the major Linux distributions such as Fedora, RedHat and Suse, Free BSD and Mac OS 10. 4+. However, you have to pick a Rocket RAID 4000 series card to get RAID 6 support with double parity.
The Green Feature
Although Highpoint doesn’t make a big fuss about it, the company now offers a green feature as well, which is meant to stop all the drives’ spindle motors to save power when they are idle. While such a feature doesn’t make sense for performance systems to ensure storage availability, home servers or small business storage systems may very well work in green mode outside office hours.
We knew about the feature, but we couldn’t find it on the management console. As Highpoint told us, it has only been implemented on the Web GUI, where we were able to turn it on. However, our Fujitsu MBA-3174RC drives would not spin down at all, so we contacted Highpoint again. We were told to use Seagate drives, as these were validated for the feature; we grabbed two Seagate Cheetah 15K. 6 drives and quickly created a RAID 0 array, but the drives again would not spin down. Unfortunately, the feature doesn’t seem to be working at all and we cannot vouch for its functionality at this time.
Performance
We ran our standard benchmark suite, which we use to test all sorts of storage controllers. In the streaming read/write tests, the Rocket RAID 2640X4 does very well in RAID 0, where it delivered the same ~470 MB/s sequential read throughput as the Adaptec 5405 card. Writes were only marginally slower, and RAID 5 performance was even superior to the numbers we received from the Adaptec controller. In RAID 10 mode, however, Adaptec implemented a more efficient read algorithm that delivers almost 450 MB/s, while Highpoint starts at 230 MB/s throughput and speeds up to 420 MB /s as queue depths increase. RAID 10 write performance is similar again.
Clearly, Highpoint can match Adaptec’s performance using four 15,000 RPM Fujitsu MBA3174RC drives. However, sequential throughput is only one half of the performance evaluation—we also have to analyze I/O performance. Whether we look at the database, file server, Web server or workstation benchmarks, Adaptec’s 5404 always provides 5% to 15% more I/O operations per second than the Highpoint Rocket RAID 2640X4. Clearly, the missing cache memory has an impact.
Could you please also benchmark software RAID solutions? I believe that these are more relevant to home/small business users who would rather spend their $350 on hard drives and use existing controller ports. Perhaps a simple Linux MDADM setup vs. Windows 2008 software RAID and compare each when used as a NAS device? I believe that you can get very good results.
Whatsmore I personally use a Linux MDADM array in conjunction with the IETD software iSCSI target to boot my Vista gaming machine (with a cheap Intel Pro/1000 PCIe adapter and Intel's iSCSI boot firmware). Could you please investigate this and compare this approach to other RAID and diskless scenarios? I have written an article on MDADM and would be happy to help you.
regards
TheDudeAbides
So how does each solution compares with ICH10R in regard to performance?
I would like to see some CPU utilization numbers. I don't see anything on the Highpoint card that resembles and currently made XOR processors. This would likely explain the degraded RAID 5 performance (most XOR processors cannot compare to a modern CPU).
I somewhat agree with theDudeAbides, but I think he knows as well as I that Linux software RAID will blow Windows out of the water. In addition, Windows failure recovery is abysmal if even possible (Windows RAID = data loss).
Speaking of failure recovery, how well did the controllers handle it? I know the initial string of LSi SAS cards drove me nuts. Adaptec has tended to be the misunderstood child on the block. Its lower level functions can be confusing to many SA's, but they are darn good. How does the Highpoint stack-up? Thanks guys for the good read.
Seriously? More 1 inch graphs, come on.........
i have the hp x4 card.. it sucks as a sata setup.
it will not go past 250mb/sec read..
4x 1.5 tb seagates or 4x300g vraptors.
iv had it for 6 months, its now also having overheating problems, iv put a fan on the side(which has fix the problem) i think it needs a heatsink? pcb gets to 50 + chip 55 rebuilding r5 4x 1,5tb setup..
im goint to test raid 5 recovery + r0 using vvs 4k sizing for xp *6tb size, to see if recovery of vvs made hdd can be done?
using file scavanger pro 3.xx (tested 2tb r0 recovery (2hdd) from this card to sil 3132.. now will try vss 3g partition setup r0 (2hdd) data recovery?
the sil 3132 does not support vss(i think) so can the software recover from the hookup as normal hdd on the sil 3132?
Where can I find a comparison between add-in RAID cards and built-in motherboard RAID?
Heya,
Software RAID solutions are much slower in the numbers than these. You can look at Tom's forums in the storage board for examples. There are ways to fine tune software RAID to be much faster though (manually setup the stripe size, cluster, sector, etc). And software RAID5 is abysmally slow even with tweaks, for someone needing performance (it's ok for serving media at home though).
As the article pointed out, these are targeted as businesses. Not home users. Why would a home user need SAS? Why would a home user be purchasing a couple hundred dollar raid controller and then put some cheap ass HDD's on there? Makes no sense.
The only reason to really get one of these, for home, would be the hardware RAID5. But unfortunately, the Highpoint card only has 4 ports, which makes RAID5 kind of stupid, as you would want 3 drives or 5 drives for optimal levels of performance (in terms of the stripe/sector/cluster; it divides into 3 and 5 drive setups just right, but not on 4 drive setups). If that Highpoint had a 5th port, I'd totally buy one for 5 tb drives in RAID5 to serve up the DVDs/Blurays camping there.
Cheers,
Personally I am also interested in the numbers if for a small home office small business set up how well the raids woirked with mainstream hard drives in the sata protacol the controler works but where does it fail in the numbers? that would be way more informative to a home office small office looking at an inexpensive raided solution also allows those juice jumbo drives you mentioned.;-)
CPU Utilization would be something interesting here as well...
Can someone fix the links please...
Some have http://http://www.tomshardware...... in them, others have the correct address but random spaces have been added.
> Software RAID solutions are much slower in the numbers than these.
Not my RAIDCore array. 5 x 1 TB WD Greenpower drives in RAID-5 (TLER enabled) on an RC5252, 250 MB sustained, reading and writing. 8.5% CPU utilization for reads, 45% for writes at that speed. This is on a 2.2 GHz Athlon X2. You'd get similar numbers with VST Pro combined with Intel's ICHxR.
The RAIDCore driver software is rock-solid, which is exactly what I want for a storage system. Highpoint drivers are garbage. Other bonuses: no heat problems, no fans, lower power use.
What's this RAIDCore missing? A few things. It doesn't have RAID-6, which is helpful when you start stacking more than 5 or 6 drives. Write cache for RAID-5 is the system RAM, which has a couple of implications:
1) You need more of it than you otherwise would.
2) The cache size is limited only by the amount of memory you have.
3) There's no battery-backup, so you'd better have the system on a UPS.
RAIDCore also isn't the best solution in the world for servers that are required to do more than just serve files. Write speeds in RAID-5 are limited only by CPU power, the speed of the drives, and the PCIe interface. If I was striping SSDs instead of these 5400 RPM GP drives, I've no doubt I could have maxed out my CPU use at 100%.
What about Intel's ICHxR RAID? That's the only decent integrated RAID solution available, but it doesn't support online capacity expansion, background array scans, or anything else I'd consider essential. Nor does have monitoring tools anywhere near the capability of the RAIDCore setup. It's also not as fast as RAIDCore. But, relative to all the other integrated RAID solutions, it's the only one that's stable. Even so, if I wanted to run RAID-5 on an Intel board, I'd still pay $50 for the VST Pro software.
What about Windows RAID? It's even more bare of features than Intel RAID. It also requires Windows Server, which costs as much as a decent RAID card. There's a hack to enable it in Windows XP, but if you're willing to trust your data to a hack that could be "fixed" with Windows update, you don't care about that data.
Linux RAID? Fast, but not as fast as RAIDCore. Expandable, yes. But administration is largely by command line. If you're a Linux person, go for it. For everyone else, there's a learning curve.
Unfortunately, Ciprico (the maker of RAIDcore and VST Pro) are bankrupt and have sold RAIDcore to Dot Hill. The link to purchase VST Pro from Ciprico's website has been dead for some time.
This is true, though Ciprico's bankruptcy is more indicative of a lack of business savvy than of anything wrong with the product. The 8-port PCIe RC5252 occasionally pops up online for $150 to $200. That's the only way to try RAIDCore at this point.
http://www.nowdirect.com/exec/search.tsb?sp_q=ciprico
The non-RAID-5 version is the least expensive 8-port PCIe controller available from anyone. It's a great alternative to the Supermicro PCI-X board that's popular, but bandwidth-limited in PCI systems.
ok, enough with all these NAS stories, how about a comparison of your top 3 NAS with a single low power HTPC and run some benchies!
sorry, my bad, i looked to quickly at the title of the article and only so the AS and thought another NAS.
Would there be any CPU utilitization rate charts with both cards?
I got this card hp 2640x4 i thaugh it would not have a bottle neck,- would be better than the onboard raid..would do me 500+mb/sec transfer (higher than my setup)... but it wont go past 250mb/sec on a setup than can get 440 usint the onboar raid..(web site says it can get much faster speed) (when i was looking to buy it.)
it seams most main chipset raids arw just as fast (r0)r5) (but not rebuilds) mas many 8x raid cards.. but have no battery backup + additional cache ram.
as a bonus, i (in future) could hook up sas hdd..
Where can I find a comparison between add-in RAID cards and built-in motherboard RAID?
That's what I'm missing too. I'm running a raid 5 on my ich9r controller, and gigabyte tells me that it varies between 70 and 260MB/sec because it's an onboard controller.
So a sandra xp startup bench might've been appropriate so we could compare with onboard stuff.