Green or fast drives for a high performance NAS?

[dkazaz]

Hi to all,

I'm in the process of upgrading from an intel SS4200-E NAS to a Synology DS1010+, mainly for the improved expandability and performance.

I'm selecting hard drives as I expect the system to be delivered in a week.

I'm torn between "Green" drives like the the Samsung Ecogreens or normal desktop drives with higher performance. I don't mean velociraptors or Caviar Black, just normal 7200rpm drives.

Ideally I'd like to have drives that run cool, quiet and burn little power (I know the difference is tiny), but I'm a bit concerned about the performance drop involved, especially since the DS1010+ is a high performance product. In tests it seems to max out near 100MB/s compared to my intel which maxes out around 55MB/s with 4 seagate 7200.11 drives.

So my question to anyone who has used these drives in a high performance NAS is, is the performance drop visible / significant? or does the fact that we're using raid 5 and doing data transfers over the network make the question redundant?

Any advice will be appreciated. Thanks!

 

[thiggins]

Many of the NASes we test now come with "green" drives. I haven't done an evalutation to gauge the performance difference between green and hi-perf drives, so don't have any hard data.

But keep in mind, that to achieve transfers above 60 MB/s or so, you need to run a dual-drive RAID 0 or SSD in your clients.

 

[dkazaz]

Really? That's kind of surprising I often see ~100MB/s disk to disk transfers inside windows from my velociraptor to a caviar black drive and back. 60MB/s is common.

I don't understand I wouldn't get it over the network (although I haven't tested the disk to disk transfers for cache effects)....

 

[thiggins]

Network file systems have more overhead and use smaller block sizes.

I'd think SATA drive transfers, even local would more often be closer to 60-70 MB/s than 100.

The Fast NAS series has a lot of info about limitations in building a fast NAS.

 

[dkazaz]

Hi Tim,

well you're more knowledgeable than me in this, but I'm surprised.


Here's what's leading me to think otherwise: For a while my Velociraptor ran in Multiword DMA 2, while the rest of my drives were UDMA5. The drive to drive performance was about 40-60 MB/s.
The problem was an incorrectly installed northbridge driver.
Once I fixed it and activated AHCI, performance from the raptor to the fast Caviar drive shot to ~100MBPs.

Now that can't be caching effects surely....

Anyway I will read through the article you mention and hopefully will end more enlightened than confused!

 

[dkazaz]

What I meant in my previous post is that since speeds from the velociraptor shot up while speeds between the other drives stayed at ~40-60MB/s, it must mean that the speeds are real and not due to caching, or it would apply to all the drives, not just the fast one....

Though it occurs to me that that being the maximum observed, speed, it could be that the 100MB/s transfers benefit from caching, while other slower transfers on the velociraptor - at about 60-70MB/s are the normal ones, meaning the 40-60 MB/ transfers from the other disks re also under caching, and some transfers are even slower.

I'm even more confused now

 

[thiggins]

I don't know everything and learn a lot from my readers. So, don't worry about offending me as long as you have data!

If file sizes you're transferring are smaller than drive cache, you'll see higher speeds than the ~70 MB/s "wall" I encountered in my Fast NAS series. The hard limit is rotation speed of the disk platter, for sequential access.

If you test with files larger than drive cache and system memory, you should see the drive rotation limits.

But, back to the original question, file transfer performance across the network is more limited to the protocol used SMB, AFP, NFS, OS optimizations and size of files than it is performance of the hard drives.

Last time I looked at this, I didn't see any significant difference among drives.

See Is A 2.5" NAS For You? and Does Drive Performance Matter in your NAS?

 

[handruin]

I've seen on many occasions file transfer speeds in excess of 110MB/sec using drives like the Samsung 1TB F3 (7200RPM) drive on large files (2+GB in size). Granted this drive isn't labeled with the EcoGreen moniker, but it's a good performing drive at a reasonable cost that runs cool and quiet (I own 3 of them). The sustained transfer rate of the 1TB F3 drive actually out-performs my 300GB Raptor drive in some cases. However, the raptor wins on seek time on random read/writes because of it's 10K RPM and smaller platters. If you want me to do some benchmarks on the F3 drives, I'd be happy to post some.

The spindle speed is one variable and the other one not mentioned is the platter areal density in larger drives. You're taking a drive like the 1TB F3 which has two 500GB platters and they're still running at 7200 RPMs, but more data per square inch is traveling by the heads when compared to my 2.5" 300GB 10KRPM Raptor. I have three of the 1TB drives and 5 of the 1.5TB drives (3 platter) which are called their EcoGreen model. The Samsung EcoGreen drives run at 5400 RPM, but they're still good performing drives.

I mentioned in another thread that I'm using the EcoGreen 1.5 TB drives in RAID 5 and I've seen 400MB/sec transfer rates in reads and 380MB/sec in writes. The seek time is obviously slower due to the reduced RPM (somewhere around 12.5-14 ms in random seeks), but so far it's been a stellar performer in my NAS. The drive array I built far surpasses my GigE network bandwidth and also that of what my NAS is capable of delivering over the network.

You'll also want to consider the RAID 5 stripe size depending on what you plan to use your NAS with the most. If you're storing large bulky files, a larger size might be ideal, but if you're planning on lots of smaller files (jpgs, mp3s, docs, etc) you might want a smaller size.

 

[dkazaz]

@handruin

Thanks a lot for your info - you've almost made up my mind for me - your other post was in response to one my threads too and I'm keen to see your benchmarks from the F2's especially!

 

[handruin]

I haven't looked into the synology DS1010+ that you're using (or planning to use). Does it have a hardware or software based RAID controller? I'll post benchies with individual drives on a normal SATA controller and with my hardware RAID. I don't want to skew your impression of the drives when using the Perc 6i because it seems to be a hell of a good controller so far. I was using a motherboard with 6 SATA ports based on the nForce 570 controller in a software RAID 5 which makes me throw up a little in my mouth when using it. The performance was ok, but I was having odd issues with both OpenFiler and freeNAS. I blame the controller for the issues because the Perc 6i has been great on the same platforms.

 

[00Roush]

Figured I would I would put in my two cents as well... When transferring files ranging from about 1GB to 20GB I usually see local speeds of around 90-110 MB/sec between drives. Transfer speeds over the network for the same file sizes is usually 85-100 MB/sec. This is using a single 320 GB WD SE16 (basically a Caviar Blue) hard drive on each side of the transfer.

While I think the current generation of green drives would work just fine I don't know how they will affect RAID 5 performance. I recall Dennis Wood mentioning better performance from Seagate Enterprise drives than from his green drives. Haven't heard much else besides that so it could be isolated or related to the older generation of green drives.

00Roush

 

[dkazaz]

Yes, I'm not surprised by those results - the seagate enterprise drives are likely to be both faster and heavily optimized for I/O, while the previous generation WD GP drives were only good at throughput and not fantastic at that.

So depending on Dennis' usage pattern, differences could be significant.

It's my understanding the newer WD GP drives with the 4k sector formatting outperform the old ones (they also have 64MB cache) and also have improved ECC capability for the capacity. However", being "bleeding edge" technology they also have increased risks and complications so I'm staying away from them and most likely opting for the samsungs.

 

[didier3001]

Hi there,

I just discovered SmallNetBuilder yesterday and it helped me a lot to understand and compare NAS products.

Great job, continue like this!

I would like to buy a Synology DS1010+ or a QNAP TS-459 Pro, I do not know yet which one I will choose but I already know that I would like to use the NAS to store and use my Virtual Machines stored on the NAS. Does anyone can confirm that high performance disks will be beneficial in this configuration.
To handle a maximum of Virtual Machines as possible, I probably need to choose a Hard drives with high I/O and low access time?

Does anyone has experience on running Virtual Machines on a NAS server?

Thank you in advance for your help,
Didier