Build vs. black box

[omegahelix]

Hello everyone. Great site...it really helps one make an informed decision. I have not been able to find an answer to my latest question, however. That is, I want to know if I would realize better performance in building my own NAS server versus buying a black box. My office already has an old school Infrant (now Netgear) ReadyNAS NV and a Thecus N5200. Both are in RAID 5 and the Thecus backs up to the Netgear nightly. The problem is we are only getting between 20-40 MBps read speeds from the Thecus via our gigabit LAN. Also, we are ready for more storage, on the lines of 7 or more disks.

My boss and I are going back an forth a bit on build vs. buy. I'm thinking we are less likely to have performance issues if we build our own and that it would also be more future proof (adding more disks, upgrading RAM, processor, etc.). My boss is of the mind that it would be easier (save my time for other things) to just get a newer black box and drop the drives in. I have also been unable to find supporting data to convince him that building our own would perform better.

In any case, I read the article "How to build a really fast NAS - Part 6: The Vista (SP1) difference" and was excited by the performance that was obtained. However, from what I can tell, you had the RAID setup at level 0 which would not work for us as we want some redundancy. Also, only one of our clients would be using Vista. The others are Linux, XP and Macintosh.

So I would like to know how building your own storage server with large (1 TB+) SATA disks in RAID 5 (hardware or software RAID, if comparable in performance) on gigabit ethernet would compare to the fastest NAS boxes such as the new Netgear ReadyNAS or Qnap or what-have-you.

If this question has already been answered, please point me in the right direction.

Thanks!

 

[thiggins]

First are the read speeds you reported in Mbits/sec or MBytes/sec. If MB/sec those speeds are not that bad.

The current crop of high-end NASes using Intel Celeron CPUs are priced relatively high compared to what you can put together. The big difference is in the software. Even the widely-used FreeNAS isn't as easy to use as the firmware that comes on the commercial NASes. And you won't get RAID expansion and migration from Linux distros (at least not from any that I know of). So your boss' inclination is spot on in that regard.

As you saw in Fast NAS Part 6, the client side of things also has a big effect on read/write speed. As the article showed, the record size used for the network file transfer will ultimately limit your throughput across even a Gigabit LAN to 60 - 80 MBytes/sec if you don't go above 64KB record size.

You also might find Intel Atom vs. VIA C7: Which Makes a Faster, Cheaper NAS? interesting. Those low-cost chipsets actually are sort of the sweet spot for DIY NASes. You can now get Atom and VIA boards with four SATA ports. Expect RAID 5 performance, however to be lower than what you see in the article, due to its higher overhead. But you might want to think about RAID 1 vs. RAID 5 for cost-effectiveness.

To sum up: When you get into the high-end NASes, you can build cheaper than buying. But the software won't be as easy to use and you'd better be comfortable with open source distros.

 

[ccb]

I'm going through the same thought process right now (build vs. buy), and I'm leaning towards 'buy' due to the software. Sure, building everything yourself might give you more versatility and expandability, but you know the old saying, "Give me enough rope...".

BTW - great site, I've learned a lot from it.

-CCB

 

[omegahelix]

Thanks for the responses. thiggins, I was quoting megaBYTES per second. I just copied a 4.2 GB file from the N5200 to my system and got an average of 17.98 MB/s on my Intel P4 3.2GHz with 3 GB RAM and gigabit LAN. It is pretty close to the 31.8 MB/s you reported for RAID 5 reads from this box, considering that others might be using the NAS right now and considering I've seen it get up to 40 MB/s at times.

I still wonder how a homemade RAID 5 or 6 box with fast processor and lots of RAM and even with a hardware RAID card would perform compared to the high-end black boxes such as the Netgear ReadyNAS Pro or Qnap Turbo NAS. I think such a test would give you a nice benchmark to compare all of these NAS boxes to.

 

[thiggins]

I still wonder how a homemade RAID 5 or 6 box with fast processor and lots of RAM and even with a hardware RAID card would perform compared to the high-end black boxes such as the Netgear ReadyNAS Pro or Qnap Turbo NAS. I think such a test would give you a nice benchmark to compare all of these NAS boxes to.

The RAID 0 test NAS that I used in Part 6 pretty well represents the best you'll be able to do. A good hardware RAID card should just negate the additional RAID 5 overhead.
I was able to break 100 MB/s (remember gigabit Ethernet is 125 MB/s), but only with the 1M record size used by Vista SP1. The limit won't be the NAS, it will be the network filesystem layer.

 

[omegahelix]

Yeah, but you used RAID 0 on the NAS server right? I can't justify the risk of running in RAID 0. If you ever have the time it would be most helpful to see what it can do in RAID 5. Of course, only if you thought it would be helpful for others too.

 

[thiggins]

I used RAID 0 on both ends, because it was the easiest way to eliminate the disk array as the speed limit.

The lesson learned from the exercise is that ultimately, for a single connection, the network file protocol will be the limiting factor, not the speed of the NAS. Throwing more compute power, either via general CPU or dedicated RAID processor won't help.