• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

pvscsi

D

defuseme2k

[H]ard|Gawd
Joined
Oct 7, 2004
Messages
1,074
I'm going to start the debate on here, if anyone even responds. Lopetive, your input would be invaluable.

I've standardized on pvscsi as my primary adapter, across the board and can find NO reason to change it. In fact, every VM uses it, we converted everything to it. There was an issue with lower i/o workloads, but according to the KB article that explained the issue -- it says it is fixed in vSphere 4.1. Its a one liner at the bottom of the KB article in the solution (how lame). If it is really fixed, something should be said, because the vast majority believe it only hurts performance except on "very" high i/o workloads.

I still see people recommending against using it at large, even respected bloggers like Jason Boche. From my own experience, I've had incredible results using pvscsi. The host CPU utilzation drops across the board; its very noticeable, but I've not found a solid way to quantify it. I've yet to see a VM perform worse afterward. Maybe we're doing it wrong.

The same goes for using VMXnet3, I've standardized on it, but it doesn't carry the same "don't use it" stigma. All stipulations apply. For instance, it isn't supported to boot linux VMs up with pvscsi, etc.
 
Do it. As you said the one case where it wasn't valid is no longer a problem...and even then it wasn't just low I/O as many people keep saying. It was low I/O with a lot of outstanding requests. The other reason people downplay it is that it used to be a hassle to switch over. All of that has gotten better.

Do it. And do VMXNET3.
 
NetJunkie said:
Do it. As you said the one case where it wasn't valid is no longer a problem...and even then it wasn't just low I/O as many people keep saying. It was low I/O with a lot of outstanding requests. The other reason people downplay it is that it used to be a hassle to switch over. All of that has gotten better.

Do it. And do VMXNET3.
Click to expand...

BOOYA. Yes, I know I oversimplified the 2000 IOs with more than 4 outstanding requests. I think that part just doesn't read into people's brains or something. I've tried to explain it, and I just get this "glossed over" look. Good thing I won this fight ;) here, but I am seeking some sort of public validation. You certainly can't find "a lot" about it online.
 
Some odd workloads don't work quite as well on it as the LSI emulated cards. If you're having no problems, keep at it :) If you get something that seems to perform "oddly" vs other guests, try an LSI card instead :)
 
lopoetve said:
Some odd workloads don't work quite as well on it as the LSI emulated cards. If you're having no problems, keep at it :) If you get something that seems to perform "oddly" vs other guests, try an LSI card instead :)
Click to expand...

Interesting, but can you explain why?

I've heard rumors that VMware is moving to make pvscsi "their" card, much like VMXnet3, it will eventually be the defacto standard. I know you probably can't say as much, but with as much speculation as there is out there, something more concrete would definitely help.
 
defuseme2k said:
Interesting, but can you explain why?

I've heard rumors that VMware is moving to make pvscsi "their" card, much like VMXnet3, it will eventually be the defacto standard. I know you probably can't say as much, but with as much speculation as there is out there, something more concrete would definitely help.
Click to expand...

From everything I know, that's correct, but it's a driver / card that is in constant development. Things keep getting better, but there have been workloads where it doesn't perform as well as the others.
 
You must log in or register to reply here.
Back
Top