Intel Xeon: Mixing stepping revisions?

Associate
Joined
21 Apr 2011
Posts
446
Hi guys

Quick question on something I'm not sure of, I have a Lab HP DL140 G3 that's serving me well running CentOS 6 x64 KVM, at the mo it has a single Intel Xeon 5160 (3Ghz Dual Core Xeon with Intel-VT) and I want to fill the second spare CPU socket with the same proc.

Can I mix stepping revisions, for example can I drop in a Intel 5160 with exact same spec except has a stepping rev of SL9RT with the existing SLAG9?

Again its for lab but would prefer it to be stable.

Thanks
 
Ok tried this and it definately matters as far as Centos 6 x64 and KVM is concerned, KVM doesn't start and you see the error: "kvm: CPU 0 feature inconsistency!" which is interesting!

Going to rebuild it with Xenserver 6 and see if it has any issues there, its no biggie as its just a lab system but interesting to know all the same.

The only missing features I could see were vnmi & flexpriority

Anyone know where these features are explained?

Thanks guys
 
Looks like it's an issue with KVM.
See here:
https://bugzilla.redhat.com/show_bug.cgi?id=477247

The best thing to do (assuming you want to stick at it) would be to probe the processors more deeply. Follow the guy's diagnosis of the problem in that thread.
Whether the incompatibility is tangible or not (feature wise) is something you need to evaluate yourself, and perhaps remove those checks from the KVM Driver with a bit of careful cutting and recompiling. The code itself is GPL and available with a quick google if you look for "vmx.c".
 
Thanks mate, I did notice that earlier when googling but didn't read it properly where you pointed out - cheers

It has a link on the comment you pointed out which shows each revision of CPU, the ones I now have are:

Intel 5160 SLAG9
Intel 5160 SL9RT

Note the different voltage, not sure what affect this will have? CentOS was running fine without any issues just not KVM, like you say I could check the source code but I'm worried about the voltage difference (even though its only a lab system) - what do you think?

I'm half tempted just to buy a single X5355 and then grab a second when I've got some free cash, but it would be nice to run it with what I've got if I can.

Thanks again for your help mate
 
Sorry I fudged it initially, it didn't occur to me that KVM might have issues with it. :(

The voltage is only an issue if the chip isn't getting enough. The variation in voltage would probably be for Speedstep and ACPI settings (C0 - C3). If you're not too worried about a little extra power consumption, you could just set it at say 1.4v and leave it at that.

The part relating to what you're interested in is here on lines 6197 to 6207:
http://lxr.free-electrons.com/source/arch/x86/kvm/vmx.c

Also relevant are lines 626 to 635, 2312 to 2326, and 2502 which is where the comparison between chips is carried out. To my eye it looks like a safety measure to stop people running two completely incompatible chips together, rather than chips of differing steppings but the same features.
Recompiling without these features might be worth a shot. It's probably worth asking over in the Linux forum!

If you're not interested in getting into the nuts and bolts of Linux, it'll be quicker to get a new CPU. Up to you!
 
Hey guys

Yamaha no big deal at all mate its only a lab system and the CPU came with the heatsink that is a pain to get hold of (I only had one) which was a big bonus, appreciate your help!

Thanks for the link mate to the code, I think I'll have a look at it and see if I can get it to re-compile with the check modified - I've written scripts but never really played around with opensource code so would be good to do! cheers

Bremen I never thought of that! good point! I'll try swapping them around tomorrow and let you know, I was trying to find out if it was possible to "mask" (disable) the extra eatures on the newest CPU that are missing on the (older revision) CPU I've just bought in the OS//Kernel somewhere so they are "compatible". (Xenserver allows you to do this apparently but only in the paid version, platinum??).

Thanks again guys
 
Ok tried swapping the procs around but it wouldnt even boot - so after a lot of messing around (thanks to you both for your help) I bit the bullet and bought 2 X Intel Xeon Processor E5345's @ 2.33GHz (Quad Cores - both identical stepping revisions) its xmas afterall! :D

If all goes ok with the new procs, I'll sell the old ones to make up the dosh and I've also got some RAM I don't need, so should almost pay for it.

Should be more than enough power for a nice lab hypervisor, what do you reckon? :cool:
 
Puts my E8400 to shame :]

Have you thought about running ESXi 5 and virtualising everything through that?
It's not quick on mine but on that it would definitely be pretty nippy.
 
Its running my website at home, the server is located in the garage (no insulation) housed on the beams in the roof, I have a Cisco aironet bridge (1310's outdoor) bridging my broadband between the detached garage and house with a 3550 cisco l3 switch in the garage routing, works well - this way if it bursts into flames it only burns down my garage and next doors and not my house lol, use Nagios Core to monitor everything - just a little project to keep my skills sharp

Using Xenserver 6.0 on my "main" (AMD Bulldozer FX 16GB RAM, courtesy of overclockers - awesome kit) system, really like Xenserver haven't used VMware for abour 2 years now - has it progressed much? was using KVM to keep up with the hypervisor market.

Impressed with Intel Xeons though, they are seriously fast.
 
The thing you got to realise about VMware is that they are hell bent on removing both Windows and Linux from the OS market and re-inventing it with there own, I don't know about you but I love Linux and I hate Microsoft (love - a Windows Server 2008 R2 server is hosting my domain & SQL Server isn't half bad) so tend to avoid it..............

Red Hat Enterprise Server + Fedora + CentOS = Awesomeness :-D

EMC.... there SAN's were never that impressive but they cost a fortune, NetApp all the way....
 
Back
Top Bottom