NASLite Network Attached Storage

Since there seemed to be a request for my benchmarks, I had some spare time today to fire up the test rig and punched out some numbers.

I used the Passmark performance software for the testing, reluctantly, I prefer to use IoZone and or diskwriggler.

The test machines are identical Dell PowerEdge 1750's with the following specs :

Dual Intel Xeon CPU 2.80gz
512KB Cache
2GB Ram
Dual Broadcom GigE ethernet
2x40 gig SCSI drives via a LSI MegaRaid in a Raid 1 config (Mirroring)

Sharing was used with SMB/Samba

A clean Windows XP Pro SP2 install was used as the client running the Passmark software. All machines were rebooted between tests to void any caching issues and all tests were done machine to machine via a *crossover* cable to avoid any network latency from cabling, switch etc.

I first tested Windows XP Pro sharing a volume. I had to test this one about 5 times because I didn't believe the numbers. Windows XP Pro had extremely low write numbers, I would have to blame this on a poor driver, which was supplied by Dell. The reads seemed pretty satisfactory and I also noticed that my copy of Windows XP was only using 1 processor, not both, I apparently didn't have the multiprocessor version/license etc.

Here's the Passmark results for Windows XP Pro to Windows XP Pro :



Next I tested FreeNAS, It performed about half as fast on reads than Windows XP Pro, but was twice as fast in writes, again the slow writes are probably due to the windows driver. (I might add, I did all the tweaks the FreeNAS site suggests for better performance)



Then NASLite, the numbers didn't surprise me, they were consistant with what I was getting during beta testing of v2. I definately believe Windows can get better writes under different conditions and of course there is slower performance since it's a RAID 1, the reads may not get better. FreeNAS has produced the same results against NASLite-2 during all my testing, and I will go on the record saying NASLite-2 is twice as fast as FreeNAS.





A few more things to consider on the server side, NASLite-2 only uses up to 768Megs of ram, and 1 processor, Windows XP Pro should have supported the full 2gigs of ram and 1 processor, FreeNAS on the other hand supports dual processors and the full 2gigs of ram.

Amen to that!

From what I have read RAID1 (mirroring) would write at the same speed as one drive but you do get the advantage for both drives on the read. At least I remember reading it in some white paper at one of the hardware manufacturers sites a long while ago. Maybe Adaptec.

Supprising results on the part of NASLite2.0. I would have put my money on Windows for the fastest transfers. This is great news in that I have a boat load of performance waiting for me in my box when I get the RAID card working.

thanks for the post Ralph,

Mike

Great post Ralph,

Finally you showed yours..

Some observations:

1. Which file system are you using on FreeNAS? The default is UFS with software update, which avoid problems if the power got cut off during operation (which seems to be a common problem with NasLite users) but is very complex and certain reduce performance.

2. A related quesiton is which file system did you have on the NasLite? EXT3 is journaling, and that will bring it on par with NTFS and UFS w/soft update. EXT2 doesn't have journaling, is much less complex, and I would expect it to perform better, at the expense of reliability.

3. Why aren't you using all the CPU and RAM? There is no point to be nice. NasLite takes over the whole system so whatever CPU and RAM you are not using is purely wasted. I don't think it makes any difference in this test. However, it will make a big difference if Naslite is used as a file server in a small business and 20 people are using it at the same time.

That brought up a marketing suggestion - you can release a higher level product with multiple CPU/core and RAM support, call it "NasPro" and charge more for it.

4. To compare apple to apple, you may want to do a test with XP w/FAT32 vs FreeNAS w/ UFS (no soft update) vs NasLite w/ EXT2, then another test with XP w/NTFS vs FreeNAS w/UFS (soft update) vs NasLite w/ EXT3.

I used the default UFS with updates. The results are negliable without updates as well.

NASLite-2 used ext3 and will fsck in the background in the event of power outage, unlike ext2

Since the bottlenecks are the nic card and the hard drive speed, how will an extra cpu and more ram make any difference?



Do you really think your going to gain/lose 10+ mb/s in any of those scenerios?

Apple to apple would mean out of the box with default settings. In the case of naslite and freenas, both are servers and should be optimised out of the box. Ralph makes a good point. There is no way that freenas can meet the performance of naslite. The items you are describing are all functions of additional CPU cycles, which by the way should have bee handled by the secondary CPU as well as buffered by the additional RAM. So what's your point?

Another thing is the naslite numbers may well be the windows client max numbers. Anyone can shed some light on that one? Ralph?



With some hardware there is nothing you can do to prevent this regardless of OS or FS. IDE does what the CPU tells it to and in many cases the CPU goes nuts when the power is cut out.

Good question!

I think it's obvious from the numbers Windows maxed out freenas, but did naslite max out Windows on reads?

Something that comes to mind at this point is that we start to run into some of the limites of the bus used. Most if not almost every one of the servers out there are based on 32bit/33MHz PCI buses. If you have the HDD and the NIC on the bus there are going to be limits hit. I think is is clear that NL is more than capable of hitting those limits. It is clear that if the drivers were up to the task, then we would all be very happy users. The 3Com and Intel cards all seem to be good performers. I at present have a netgear GA621 fibre gigabit card and it seems to work great though I am no where neer the limits of the the card. To exercise it proper I need to get my RAID card working in it.

Now the real question is what NICs have really good and stable drivers that are going to allow us to leverage the performance that is obviously there to be had in the OS? Also what RAID cards have good, stable drivers and will perform as well. I know that Ralph uses LSI RAID cards. I also know that there are a good few that use 3Ware cards as well. Finaly there is the Switch performance issue. I can't see how a switch would cause the serious slow downs noted but then again I use only really good managed switches. I guess that there might be a need for a list of switches and chipsets that performwell under the stress that the NAS box is capable of putting on it.

I will leave it at that as I think that there is more than enough to ponder and get answers on.

Mike

To compare apple to apple - XP should not even be considered, since it is a client OS - ideally Windows Storage Server should be used, since that is Microsoft's storage optimized offering, which - for those of us who may not be aware - is not licensed on a per user basis.

I didn't know about that offering from MS. Still don't matter to me though. The MS option would likely perform not much better and would cost much more. Also the setup would be stupid hard compaired to that of NL. It would allow much more configurability though and user control. Then again if I were in the market for that kind of control and power, I would be going to Sun and just DL the software for free from them. The added plus of the offering from sun is that it is a unix based OS.

Mike

i would really love it if you could add naslite 1.x, so we can see performance gains in updating

It is likely that the performance gains would be minimal if you were just talking about non RAID transfers in 1.x or 2.x. Add in the RAID and NL2.x will whip the crap out of 1.x.

Mike

Also another thing to consider is scalability. I bet that naslite will be able to handle far more users on given hardware than any other OS.

As I said, it won't make any difference in this test. However, for servers, single connection test is very limited. Most users use NasLite at home so it doesn't matter, but small business users may have dozens of users accessing the NAS concurrently and I believe it will make a difference - at least the RAM usage will.

Keep in mind, the file system structure is loaded in memory. It doesn't matter much in normal operation, but if there are millions of files, the file system itself may use up all the RAM and cause the server to die. I have heard stories this happen on 2GB RAM machines. Unusual situation though.



Since 2.x has the web interface, it uses more RAM to run Apache and PHP. That could may a difference if the RAM is limited (64MB? 128MB?)




I will have to disagree here.

In the web hosting circle, it is well known Linux is faster than FreeBSD (this may well be the reason NasLite performs better than Windows and FreeNAS in this test). However, FreeBSD is the more stable under heavy load. At certain point (dozens, or hundreds concurrent users), all NAS OS will have difficulty to handle the load, but without testing, nobody knows when that happens. The RAM usage mentioned above may well become a problem for Naslite under heavy load.

That's why I suggested to have a NasLite Pro (with multiple CPU support and higher max RAM usage) targeting small businesses.