RAID1 (mirror) - performance - mac mini

Question

RAID1 (mirror) - performance - mac mini

Hi All,

I thought I'd share my specs/info on performance of a mirrored mac mini... mostly because it's horrendously slow. 🙂 I'm new at OSX Server, so it will be good to get some feedback.

Specs:
OS: 10.6.6
Proc: 2.66 GHz Intel Core 2 Duo
Memory: 4 GB 1067 MHz DDR3
HD: 2 500
- RAID block size: 128K
- Mac OS Extended (Journaled)

Network: Gig Switch.. currently on 100basetx because of economy cat5 cable.

O.K., so I finally got this MacMini server configured and working within my network.

I first go to transfer some files to it (large transfers of a gig+ at a time). It appears I'm getting a transfer rate of about 25MB/min.

Of course, this frustrating. I then realized that my nic's were set at 100basetx and not 1000base... This appears to be because of the cheaper cat5E cables I'm using because I tried some Belkin cables which then sparked the 1000base t full duplex mode.

I was very hopeful with the Belkin cables! Anyway, this did speed things up, but only slightly. I still was getting slow transfer rates.. but maybe around 50MB/min.

This was not going to work for me having to transfer 400Gigs of data.. so I decided to start my client in Fireware 800 target mode to mount it directly to OSX Server. I was thinking up to 800Mb/s would get things done fast. What I came to find out was that transfer rates were still slow. The lightbulb went off and I finally figured it was the Mirror that was slowing things way down.

(btw, the client, in target mode, connected to the server directly had some strange behavior.. after some transfers, the mounted volumes would appear to duplicate (on the Desktop of the server).. then it would not allow me to copy once that happened. It would throw errors. The only way to unmount them was to shutdown the server (as I didn't want to force unmount them). Strange!, but no loss of data)

Anyway,
I realized initially that a mirrored drive would make for hindered performance, but I never thought it would be this hindered.

That is the skinny.. thoughts?

Donovan

MacBook Pro, Mac OS X (10.5.8), MacBook Pro, iMac intel, iMac G3 (webserver), Powerbook G4, MacMini G5, others

Posted on Jan 14, 2011 12:37 PM

Reply

Answer 1

Jan 14, 2011 1:42 PM in response to Donovan Brooke

Mirrored drives will give you 120% of normal performance during reads, vs 80% for raid 5, but, like any raided device will be a bit slower during writes because both discs need to be written to.

I doubt the problem you are having is the mirrored drives. First, you need cat6 to run GigE (1000BT), otherwise you will end up with massive retries and thus slower transfer. Buy the cables. Second, make sure that your GigE switch is not one of the cheap $50 varieties.

All switches contain some amount of shared ring buffer memory which is used to buffer data transfers between ports. Cheap switches tend to put in about 128K shared between all ports and when it's out it can do nothing but flip the "stop transfer" switch on until it has something free. You need to be looking at, for example, an HP switch for around $140 that has 128M-256M of memory for GigE work which allows the switch to move massive blocks of data without constraint.

Reply

Answer 2

Jan 14, 2011 2:30 PM in response to Donovan Brooke

Hi

I doubt very much if the RAID 1 Mirror is at the heart of the slow down you see? I've deployed quite a few MacMini Servers since they came out. Always with a RAID Mirror. None of them suffer from the performance you seem to describe.

I think what you're seeing may be due to a misconfig? Probably DNS? You've not said what services you're running. Until you eliminate that possibility it's may be difficult to troubleshoot?

Tony

Reply

Answer 3

Jan 14, 2011 4:23 PM in response to Antonio Rocco

Hi Guys,
Thanks for the comments and good suggestions.. only those don't address the fact that I was connected directly via firewire with slow performance as well.

I did some more tests.. I simply copied locally:

9.7GB (misc. data to a new folder)

This took 12 minutes to finish, which in my calc:

800MB/min, or 13.3MB/Sec.

Is it just me or is that slow?

Donovan

P.S. I'm running most of the services with no load at the moment.. since this server is set up on my LAN.

Reply

Answer 4

Jan 14, 2011 4:26 PM in response to Donovan Brooke

correction, it appears I'm getting about 150Mb/Min transfer.. but that still seems slow. - Donovan

Reply

Answer 5

Jan 15, 2011 2:51 AM in response to pcolvin15

Just to add: depending on cable length and interference from other sources, Cat5E will properly do gigabit ethernet even if it is not certified to do so.

http://www.networkcablingdirectory.com/articles/structured-network-cabling-id_11 51.htm

Most Cat5 you buy will be Cat5E, but watch out: true Cat5 cannot handle gigabit.

Reply

Answer 6

Jan 15, 2011 3:35 AM in response to Gerben Wierda

You are correct, Cat5E will do GigE in controlled environments where everything is clean, such as a data center. Cat6, which is better shielded and certified for GigE, is the choice when your don't have total control of the environment or want the cleanest pathway.

Reply

Answer 7

Jan 15, 2011 12:38 PM in response to Donovan Brooke

If anyone is currently running the MacMini in a RAID1 config (using the two internal 500GB drives), can you please do the following test for comparison?:

Simply do a local copy of misc. folders/files to a new directory.

I would suggest copying up to 10Gb's if you can.. if not, just do
what you can but note the amount of info being copied.

Measure the time it takes (using a clock, not the time listed in the status bar).

My live example:
9.7Gb
12Min's

9.7 / 12 = .803 Gb's/min
or 800 Mb's/Min.

It would be nice to know what other folks are seeing for local copy times.

TIA,
Donovan

Reply

Answer 8

Jan 15, 2011 2:01 PM in response to Donovan Brooke

For this test, I'd expect to get the read speed of one of the disks (best case), and the aggregate write speed to each of the two disks, less driver and file system and controller overhead and the seek times.

And this test is inherently going to generate contention; the drive heads will be busy seeking.

Try the same tests out the FireWire 800 and in from the FireWire, just for grins.

The older Mac Mini I checked was running with the HLGS 5400 RPM SATA drive, with 11 and 13 ms seek times for read and write, so far from the fastest drives around. Flat out, they're rated to show 540 Mbps from disk to cache, and I'd expect the write speed to be rather lower than that. I don't have the Intel ICH-7M specs handy, or whatever that box is using as a bridge chip.

And the contention is almost certainly going to obliterate the bandwidth here.

Computers are a system and a balance of parts, too; there are always multiple parts involved in the design, and over-designing any one piece of a computer system usually serves to move the bottleneck somewhere else, and to waste money. All of this disk I/O is massively faster than that 100 Mbps Fast Ethernet, for instance. These disks should be more than capable of keeping that pipe stuffed full until the cows come home.

And no, I don't have a matching configuration handy where I can run this test; sorry.

Reply

Answer 9

Cedari

Level 1

0 points

Aug 4, 2011 7:21 AM in response to Donovan Brooke

I have a similar setup and are experienceing the same results on a mac mini server (late 2010 model all ALU) with two internal 500 GB discs. My problem is only when using the disks in a raid 0 or 1 or when copying files from one internal disc to another.

It is panstakingly slow and the disc trasfer rate is well below what could be expected.

Same file copied from one internal disc to anoter is five times slower than copied from an external USB drive!

As I see it it might be a disc controler missconfiguration or something worse, hardware problem!

I wish Apple would look into this!

I will give you an interesting comparison, two images taken from the activity application first image is when I copy a file from one internal disc to the other. The second image is when I copy the same file from an USB disc to an internal disk

I am more used to the profile of the second transfer, here the reading (green) is the limiting factor, the internal disc is able to write faster but has not enough (starvation) data to write.

The first image (eg internal transfer) is another matter alltogether, It seems to be some kind of race condition or bus collition matter.

Reply

Answer 10

Aug 5, 2011 5:26 AM in response to Cedari

RAID 0 is striping, and there effectively aren't two internal disks in this configuration, and no data redundancy.

RAID 1 is mirroring, and there effectively aren't two internal disks in this configuration, and data redundancy.

If you want to test operations within the box and between the two disks, there can be no RAID in use on the disks.

If you want speed and data redundancy, then you'll be looking at a Mac Pro, and potentially at hardware RAID, or at an outboard FireWire800 storage device with hardware RAID, or analogous solution.

And if you think you have a hardware problem, run the hardware diagnostics and then check with Apple directly.

Reply

Answer 11

Cedari

Level 1

0 points

Aug 5, 2011 11:13 AM in response to MrHoffman

Almost everything you write is true and known, at lest by me! What is not true is that it is possibe to performance test RAID systems.

I have reinstalled the system and even used the CMD-OPT-P-R reset and I still have the same problem the system is even unresponsive it takes over 2 (TWO) minutes to start Safari sometimes.

I have a new image for you to analyse:

This is the a complete copy of a 1 GB sigle file, stored with as few segements on disc I could muster, e.g it is created with mkfile. It is done on a RAID 1 configuration.

It should not be so spiky, what is the system doing in the vallies? Why can it not read at a constant spped as shown in my earlier post, second image.

I am gonna send in my hardware to let them look at it at ny local Apple store

Reply

Answer 12

Aug 30, 2011 9:55 AM in response to Cedari

Cedari, Interesting pic of activity monitor. I stopped worrying about my slowness issues as decided to put the machine in a non-production environment. However, I'm interested to see what you find out, as I never did figure out why the system was so slow. I've worked with quite a few RAID1 setups and performance was never this bad.

Donovan

Reply

Answer 13

Cedari

Level 1

0 points

Sep 16, 2011 12:28 PM in response to Donovan Brooke

Glad I could be of help!

According to my Mac Store rep. One of the disks was defective and was replaced. The machine has become alot faster but not as fast as my older version of Mac mini (not server). It bums me out that I got this new Mac mini server with raided discs and it is slower than my old machine.

I am leaning towards trying to replace the other disc too. That might speed it up even more.

Reply