Is this hardware acceptable or should I swap out the MB / CPU / RAM ?? Super Micro -- Is this ECC?

[danb35]

On the topic of the i3 that has to be a particular model with the 'e' on the end of the part number to support the ECC ram. Is that correct ?

That is not correct. Pick your preferred i3 and it should work; you can check on ark.intel.com to confirm that it supports ECC.

 

[Ericloewe]

Ok, I'll look into the I3 then. Yeah I currently have an M1015 but I'd need a second one. The reason I didn't consider that board was it looked like it only had Sata connections no SFF-8087 connectors for the backplanes in my chassis. I was going to get a second M1015 as part of this upgrade.

On the topic of the i3 that has to be a particular model with the 'e' on the end of the part number to support the ECC ram. Is that correct ?

Could you possibly post a link to a suitable i3 for me?

Thanks,

As jgreco said, use reverse breakout cables and save some cash by going with the X10SL7-F.

As for the processor, no, not at all.

All desktop Haswell and Ivy Bridge i3s support ECC:

http://ark.intel.com/search/advance...ors&MaxTDPMin=50&MaxTDPMax=300&ECCMemory=true
http://ark.intel.com/search/advance...ors&MaxTDPMin=50&MaxTDPMax=300&ECCMemory=true

I've filtered out the low-power ones, because they don't use less power at idle - they just end up performing worse when you do need them to work.

 

[antsrealm]

Ok that's some excellent advise guys, really appreciated.

So the i3 seems like the go with the X10SL7-F utilizing the reverse breakout cable. (Was not aware of that cable).

The only thing really that's concerning me now is the amount of RAM.

Everyone says you need a lot of RAM with FreeNas and allow 1GB for every TB and then some. So if I have in total 16 3TB drives with 4 of those being redundant wouldn't I need at least 48GB of RAM. With these boards only going up to 32GB max I have no where to go if it turns out it's not enough.

Some people have indicated this will be ok with 32GB of RAM but I really need to be sure before I spend the money to find out it all needs to be replaced again.

What's your thoughts on the RAM situation ?

Thanks,
Tony.

 

[mjws00]

tl;dr You asked the RAM question. ;) Eric's answer "you should be fine." is on target.

A little more logic. With ZFS ,RAM is used primarily as ARC, which is your primary cache. So the idea is that on frequently accessed reads, we can serve directly from memory and not bother hitting the disks. The intent behind 1GB per TB is that we scale up ram as the pool grows, so that we can store a similar percentage of our reads. This lets us hit the cache (x)% of our reads for a given workload. Remember it is assumed that as that pool is scaling, users, and the number of accesses are also scaling. It is assumed that we NEED to hit the cache as often as possible, and at the very least in the same ratios. With lots of users in a corporate environment this is valid. Home users completely break that model.

How is a home user different? Pretty much the only thing that takes this much space is media. Typically we write once, read once... let it sit for months at a time. Then we access the files randomly. No hope of a dumb cache finding a useful pattern. In addition the files are huge and sequential, and our pool is MUCH faster than the network. So we aren't really using ZFS for it's cacheing properties, primarily it is being used for reliability, and redundancy. Thus we can relax the "rule of thumb". The reason we can't give you a hard guarantee is that your specific workload and usage is unique (but common). It is easy to create a workload that will beg for all the RAM you can afford... even on a small 1TB pool.

Ironically the 32GB limit is probably what pisses me off the most about Haswell E3's. ;)

 

[antsrealm]

tl;dr You asked the RAM question. ;) Eric's answer "you should be fine." is on target.

A little more logic. With ZFS ,RAM is used primarily as ARC, which is your primary cache. So the idea is that on frequently accessed reads, we can serve directly from memory and not bother hitting the disks. The intent behind 1GB per TB is that we scale up ram as the pool grows, so that we can store a similar percentage of our reads. This lets us hit the cache (x)% of our reads for a given workload. Remember it is assumed that as that pool is scaling, users, and the number of accesses are also scaling. It is assumed that we NEED to hit the cache as often as possible, and at the very least in the same ratios. With lots of users in a corporate environment this is valid. Home users completely break that model.

How is a home user different? Pretty much the only thing that takes this much space is media. Typically we write once, read once... let it sit for months at a time. Then we access the files randomly. No hope of a dumb cache finding a useful pattern. In addition the files are huge and sequential, and our pool is MUCH faster than the network. So we aren't really using ZFS for it's cacheing properties, primarily it is being used for reliability, and redundancy. Thus we can relax the "rule of thumb". The reason we can't give you a hard guarantee is that your specific workload and usage is unique (but common). It is easy to create a workload that will beg for all the RAM you can afford... even on a small 1TB pool.

Ironically the 32GB limit is probably what pisses me off the most about Haswell E3's. ;)

Ok Thanks mjws00 :) I know Eric mentioned it but I was after a second opinion and a bit more of an understanding. Your post has answered both of those needs. Ok guys, thanks for all the help it's been really appreciated. I think it's time to make some hard choices on how much I should spend :p

One thing with the ram and the super micro boards, when I go to the product page on the supermicro website and I look at tested memory I see a lot of Hynix etc but then I see people using Crucial RAM. Why is that ? I'm trying to understand how to be sure I'm choosing the correct ram brands and then whether to get registered ECC or buffered ECC or just plain old ECC?

 

[Ericloewe]

This is the reason.

Crucial is typically a lot easier to find, whereas Samsung and (especially) Hynix are harder to find.

You'll want unbuffered. LGA 1150 processors only support Unbuffered DIMMs.

 

[antsrealm]

Thanks Eric, looks like I should of read all the stickies Feels like I have now :p

 

[antsrealm]

So this is the final hardware list to go in the original chassis.

X10SL7-F-O
i3-4370
CT2KIT102472BD160B (Crucial 2 x 8GB ECC RAM)
CT2KIT102472BD160B (Crucial 2 x 8GB ECC RAM)

 

[mjws00]

Looks nice. I'll pm you the shipping address. ;)