nMP running 24h/7 ? Really Safe?

[lemonade-maker]

Call it what you want. It was a bit of a surprise to me how well the Mini can handle a load like that. In the past year, my Mini has pumped more than 3.5TB of traffic through it's Ethernet port.

The machine averages between 5-6 web requests per second (that's 14 million requests every month, for those who are counting), and I've seen it handle 10 times that many requests during high-load events without breaking a sweat. I've attached below a graph of web request rate over the past month. Gaps are due to services being load balanced with another server.

About 10% of the web requests are for dynamic content. The rest is static, which can make a huge difference. However, the content this Mini serves are images that it self-generates every 10 minutes. So while the content is technically static, it changes often and the server does all the processing to make the changes.

I also attached the Mini's load average plotted over the past year. It's definitely keeping busy…

So while a Mini isn't necessarily your traditional server (and certainly not giving you any form of redundancy if one of its components fails), it can definitely handle a good amount of load. Throw 2-3 Mini's together in a failover or load balanced configuration and you have a nice redundant server setup at a fraction of the cost of more traditional rack mount hardware (I have those types of servers too).

Thanks. That is really awesome. Why do you use hardware for that rather than a cloud solution like AWS? BTW the bullshirt was for using a mini as a prod server not your traffic.

 

[christarp]

I've been re-encoding my entire Blu-ray/HD DVD library (almost 1000 discs) on my nMP using HandBrake since February 15. I queue up my encodes so my CPU has literally been maxed out 24/7 for that entire time (excepting a couple of restarts for software updates and the like). I haven't had a single problem with this use case, so I conclude that these computers are designed for heavy, constant workloads. I still have a few weeks to go before I'm done, but I should be done with the Blu-rays this weekend!

I did this exact same thing but with a computer and a rMBP... The computer was operating at 120c+ for the cpu temperature (yikes!) but ultimately I didn't care that much about it, went for a good week like that (and it still works great! Kinda surprised it didn't keel over and die to be honest) Retina macbook pro was doing another set of movies at the time and I had it up for about a week straight as well, but that was throttling itself at around 95c. For something dedicated to professional workstation work like the mac pro i'd expect you'd be fine.

 

[mikepj]

Thanks. That is really awesome. Why do you use hardware for that rather than a cloud solution like AWS? BTW the bullshirt was for using a mini as a prod server not your traffic.

The workload I have requires a lot of memory and fast I/O. In the past, I looked at VPSes, but they were costly in the types of configurations I needed. Same goes for Amazon's server instances (though I have tried out one to see how it would do). I found it was cheaper to buy my own rackmount servers and just pay a flat collocation fee. The last rack server I bought, and still have in production, has 24GB of RAM and 15k RPM SAS disks in a RAID setup.

I always heard about the Mini being used as a server, and the price was certainly right. I could buy and host 2-3 Minis at a place like Mac Mini Vault for less money than a single heavy-duty rackmount server at a local data center.

So I gave it a shot, and was pleasantly surprised at just how much abuse these things can take. Just fill the thing with RAM and configure it with SSDs, and you are good to go.

Like I said before, I would never rely on just one for a production environment. But buy a few and use DNS failover, and it can make for a nice setup. I still haven't decided if I'll replace my oldest server with a Mini again, but it's definitely a consideration. The only thing I might change next time around is the OS. OS X Server is easy to manage for simple setups, but kind of a pain when you want to customize it. I might give Linux a shot if I can get it working without rEFIt, or maybe VMware ESXi with a couple of Linux VMs.

----------

Oh, and one other concern I had was going for consumer-grade SSDs in a high I/O workload production environment. Was wondering how long the drive would last before failing. Buying AppleCare made resting a bit easier (and is a lot cheaper than maintenance contracts for enterprise servers). I figured this way I would be guaranteed at least 3 years of use out of the box.

So far, my concerns have been unfounded, but I have another 6 months before AppleCare runs out. Maybe I'll be back here 7 months from now complaining about a failed SSD... ;-)

 

[fabric17]

Thanks everybody for the hints!
So I 'll run 24/7 , without worries!

Remains a couple of doubts:
1) Heard rumors about HI temp (> 95 C) for long time not safe for transistors
2) is SSD as reliable as HD? (with large numbers of READ/WRITE operations?)

 

[mikepj]

Thanks everybody for the hints!
So I 'll run 24/7 , without worries!

Remains a couple of doubts:
1) Heard rumors about HI temp (> 95 C) for long time not safe for transistors
2) is SSD as reliable as HD? (with large numbers of READ/WRITE operations?)

I would expect Apple designs all of their computers to run within typical thermal boundaries. As long as the environment temperature is in the specified range (in the Mac's specs), it should be fine.

I don't think SSDs have been around long enough for anyone to really have a solid comparison of how they behave with respect to hard drives. In my experience, I have never had an SSD die on me, while hard drives die all the time. I would expect this to be the case, because SSDs don't have any moving parts to wear out like hard drives do.

However, SSDs have a limited number of write cycles. You can only write to the same block on an SSD a certain number of times before it will no longer take another write. SSD manufacturers have compensated for this limit by over-provisioning storage (including more space than is actually specified in the drive size, so if some areas wear out the drive just stores the data in the "extra" space) and wear-leveling (writes to an SSD are mapped to different locations, so no one block is getting written too many times).

And again, with SSDs only being popular over the past 5 years or so, it's hasn't been long enough for anyone to really get a feel of their failure patterns (how or when they are most likely to fail).

 

[goMac]

Thanks everybody for the hints!
So I 'll run 24/7 , without worries!

Remains a couple of doubts:
1) Heard rumors about HI temp (> 95 C) for long time not safe for transistors
2) is SSD as reliable as HD? (with large numbers of READ/WRITE operations?)

I wouldn't worry about SSD read/write cycles. A lot of people like to talk about it like it's a huge concern, and don't get me wrong, it is a real thing. But SSDs are much more durable than traditional hard drives. By the time you hit that many reads/writes, a spinning hard drive would likely have failed anyway.

So I think concerns about SSDs are generally overblown.

I've heard of people buying hard drives as their primary boot drives for Mac Pros, because they want to treat the SSD as some precious delicate thing that they don't want to throw any read/write cycles at, and to me, that's just insane. You might as well buy a sports car and pull it around with a horse to avoid damaging the engine.

 

[MH01]

Should be fine.

Just to be clear these are not missions critical servers, so while they are not built for 24/7, odds are they will be just fine.

In all honesty, they have just been released so the reliability will be tested in the months to come.

Problem with the new design, once something goes wrong, you have to take it in to get fixed. The old ones, you could replace components.

Get Applecare to be safe.

----------

I would expect Apple designs all of their computers to run within typical thermal boundaries. As long as the environment temperature is in the specified range (in the Mac's specs), it should be fine.

I don't think SSDs have been around long enough for anyone to really have a solid comparison of how they behave with respect to hard drives. In my experience, I have never had an SSD die on me, while hard drives die all the time. I would expect this to be the case, because SSDs don't have any moving parts to wear out like hard drives do.

However, SSDs have a limited number of write cycles. You can only write to the same block on an SSD a certain number of times before it will no longer take another write. SSD manufacturers have compensated for this limit by over-provisioning storage (including more space than is actually specified in the drive size, so if some areas wear out the drive just stores the data in the "extra" space) and wear-leveling (writes to an SSD are mapped to different locations, so no one block is getting written too many times).

And again, with SSDs only being popular over the past 5 years or so, it's hasn't been long enough for anyone to really get a feel of their failure patterns (how or when they are most likely to fail).

hmmmm I remember getting one of the first unibody MBPs, the suckers would black screen of death as the temps got too high. General rule with Apple computers, the Revision A has some issues Get the Rev B where is been sorted and never acknowledged !

Though they built these for the average user in hand, the issue I experienced was while gaming, so edge cases, one could argue.