ECC vs non-ECC RAM and ZFS

Status
Not open for further replies.

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
So C222 > C226 PCH? Or are the differences minute.

Also for 7.2k 4TB hdds, SAS is $100 more / harddrive... which I don't think warrants the price increase for virtually the same drive with the same cache. Or am I wrong?

Fortunately you are wrong. Not about the price, but at a meta level.

An SAS controller is perfectly capable of handling a SATA drive. The inverse is not true, however.

So getting an 8 port SAS controller onboard is a pleasant bonus.
 

TXAG26

Patron
Joined
Sep 20, 2013
Messages
310
Fortunately you are wrong. Not about the price, but at a meta level.

An SAS controller is perfectly capable of handling a SATA drive. The inverse is not true, however.

So getting an 8 port SAS controller onboard is a pleasant bonus.

I'm using regular SATA Seagate NAS 5,900 rpm drives off of a LSI SAS2308 and it works great. I even threw in a SAS expander between the drives and 2308 for good measure and have noticed no drop in speed. I'm just on gigabit, so didn't feel that 7.2k rpm drives were necessary since even one or two newer 5900 rpm drives are capable of saturating a 1gb network link on their own.
 
Last edited:

IonutZ

Contributor
Joined
Aug 17, 2014
Messages
108
You can't write faster than 140MB/s on it, how does it saturate a 1gb network link??

^Scratch that I feel dumb. 1gbps = 125MB/s

Wow I didn't even realize how big of a bottleneck the network is then... and going to 10gbe is ridiculously expensive :|
 
Last edited:

TXAG26

Patron
Joined
Sep 20, 2013
Messages
310
Gigabit gear was the same way back in 2000-ish when it came out and by 2003-ish, the price gap had closed to about a 75%-100%!premium per port.

I think most 10gb switches now are running 7-10x what a similar 1gb switch is going for.

10gb has been out a couple of years already and the prices are starting to drop, but not as fast as we're accustom to.

On a positive note, many Supermicro boards are now coming with dual port Intel 10gb ports and the cost of those boards doesn't seem to have increased much compared to what previous similar models retailed for. Hopefully HP follows Netgear & Dlink and comes out with a line of quality L2 midrange SOHO 10gb switches.
 

Knowltey

Patron
Joined
Jul 21, 2013
Messages
430
I love that I've managed to corrupt a generation of NAS geeks into IPMI laziness, oops, I mean IPMI goodness.

So I did get a board with IPMI. Is it basically just I access the IP the IPMI port gets in a browser or something?
 

panz

Guru
Joined
May 24, 2013
Messages
556
Connect an Ethernet cable to the IPMI interface (read the manual to identify it) and use a web browser or the IPMI program that is shipped with your motherboard (and update it online as soon as possible). The workstation you're going to use for this purpose has to be on the same subnet of the IPMI interface.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Connect an Ethernet cable to the IPMI interface (read the manual to identify it) and use a web browser or the IPMI program that is shipped with your motherboard (and update it online as soon as possible). The workstation you're going to use for this purpose has to be on the same subnet of the IPMI interface.

Probably only if you're using discovery. It's been a long time since I've bothered with the installable IPMI tool so I don't recall exactly.

If you've configured the IPMI correctly (via DHCP, or static config) it'll happily work across subnets. Has to, else it'd be useless.
 

Knowltey

Patron
Joined
Jul 21, 2013
Messages
430
Probably only if you're using discovery. It's been a long time since I've bothered with the installable IPMI tool so I don't recall exactly.

If you've configured the IPMI correctly (via DHCP, or static config) it'll happily work across subnets. Has to, else it'd be useless.

Do I have to configure it in the BIOS prior to use or such?
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Do I have to configure it in the BIOS prior to use or such?

You need some way to bootstrap an IP into the IPMI. There are three common ways:

1) You can hook up a keyboard and monitor, and there's usually some option somewhere to set up the IPMI networking information (IP, subnet, gateway).

2) You can use a DHCP server. This has several subclasses:

2a) Use a DHCP server that is capable of informing you where it has assigned the IPMI.

2b) Make a static DHCP assignment

2c) Troll around your DHCP assigned range (possibly reducing the size of it) with a web browser looking to see where it is - the sucky option

3) Use the IPMI management tools from a PC located on the same subnet, there'll either be a broadcast or search function (don't recall).

From a service provider point of view the correct deployment strategy is to DHCP-assign the IPMI a fixed IP address (which allows you to set it up in DNS, and use proper SSL certs) and then turn OFF DHCP in the IPMI once you've made the connection, assigning a static IP address within the IPMI, so that if your DHCP server is unavailable, your network will come up in a known state. Also if the IPMI gets wiped (upgrade etc) it will come up in a known location because it will fall back to DHCP.
 

gpsguy

Active Member
Joined
Jan 22, 2012
Messages
4,472

Knowltey

Patron
Joined
Jul 21, 2013
Messages
430
Okay, the IPMI interface was going to get hooked up to the router, which has DHCP and I can easily find out what IP got assigned. Once I get the MAC I plan to do a static lease to 10.1.1.10 as well, but basically as long as I can find out what DHCP address was assigned to the IPMI interface it should be as easy as accessing it in the browser to configure stuff?.

And good idea gpsguy, I could always stick it on a VLAN that uses another subnet or such and then restrict traffic accordingly.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Okay, the IPMI interface was going to get hooked up to the router, which has DHCP and I can easily find out what IP got assigned. Once I get the MAC I plan to do a static lease to 10.1.1.10 as well, but basically as long as I can find out what DHCP address was assigned to the IPMI interface it should be as easy as accessing it in the browser to configure stuff?.

And good idea gpsguy, I could always stick it on a VLAN that uses another subnet or such and then restrict traffic accordingly.

Most home users are going to be just fine merely making sure it isn't exposed to the Internet (i.e. behind NAT/firewall) even if it's on their shared home IP range. Certainly if you understand networking and are comfortable doing so, putting it on a separate network is just dandy too.

What happens next basically depends on what the software wants. If you're using the Web based interface, for example, you will probably need Java, and sometimes the versions are finicky. Also you may run into issues with Java not wanting to use http://, so you can turn on https:// and accept the certificate mismatch, and/or twiddle Java security settings for that one specific address. If you have trouble with one browser, try another. This is just some general stuff that comes to mind from years of the frustration and joy of IPMI. Most problems are due to the fact that ORACLE SUCKS and JAVA SUCKS but it is all worth it in the end because you can manage your system remotely.

Even without Java, basic functions like turning the server on and off, checking health status, etc. will be available. But the win is typically the client that not only allows you access to the console, but also allows you to plug in devices "remotely" over the network. You can make a CD (or CD image), USB device, floppy, etc., appear as a USB device to your host. Once you've got IPMI ethernet installed and configured, there's little reason to ever manually attach a temporary storage device to your server ever again. Burning a CD? Why bother. IPMI in and attach the image. etc.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Correct. IPMI monitoring of the system event log for the win...
 

John Doe

Guru
Joined
Aug 16, 2011
Messages
635
Hallo everyone,

thanks ALOT to cyberjock for his evangelization about ECC memory. To be honest I was lucky because using old 4core xeon core2 not used anymore as workstation in my labs as 6 disks Freenas box. They all had ECC ram by default.

I read with attention everything written, the only part I it does not sound to me is what THEORETICALLY might happen in the case (RARE, IMPOSSIBLE, ONCE IN A BILLION YEAR) of more than two bits flip. What cj was saying is that the system will halt for sure, as far as I remember ECC hamming code is a SECDED (abbreviated from single error correction, double error detection) so more than 2 bits might or might not get undetected.

Am I missing something?


Thanks,

Alessandro
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
In order to get a double bit error by undetected, you'd have to have a third error somewhere in the Hamming code that itself wouldn't get detected. I'm not an engineer, but I think the odds on that are somewhere around "would occur just before the heat death of the universe."
 

John Doe

Guru
Joined
Aug 16, 2011
Messages
635
Ciao HB!

yep statistically I am sure it will never happen. I am just thinking about a somehow failing area of the ECC giving three bit error or any weird impossibility. In many posts by CJ he assumes that more than three errors are always detected and I was wondering if my info about that are outdated.

All my
 

pjc

Contributor
Joined
Aug 26, 2014
Messages
187
Correct. IPMI monitoring of the system event log for the win...
Does this just entail running "ipmievd sel timeout=60" or is there anything else I need to do?
 
Status
Not open for further replies.
Top