151

Re: Meltdown and Spectre

Hello, Evgenie Muzychenko, you wrote: the C>> Then then is necessary all TCP and other to implement in userspace. I eat> What for? If it would be desirable to implement direct dialogue with a network card. Sockets and ring  somehow together not especially live.

152

Re: Meltdown and Spectre

Hello, lpd, you wrote: lpd> Calls to user-mode a host all the same transit through a host kernel. What is "calls to user-mode a host" and for what it by all means to transit through its kernel? lpd> on mine  when VM (compilation) is loaded on 100 %, virtual machine process on a host loads percents of percent on 120-150. I.e. on a host many operations are fulfilled. What sense to look at processor loading? It will be always loaded by the active operation, differently it is impossible. Look frequency of switchings in a kernel.

153

Re: Meltdown and Spectre

Hello, Cyberax, you wrote: Cs> Sockets and ring  somehow together not especially live. What exactly could prevent to live it together?

154

Re: Meltdown and Spectre

Hello, Evgenie Muzychenko, you wrote: I eat> I eat>> what  in the settled operation mode  thousand-ten thousand calls to a host kernel? I eat> S> On which are twisted , for example. I eat> It is possible an example typical (that is, rather spread) a DB which the majority of [tens thousand requests in a second needs to be led in the transactional form? In my opinion, the most part of requests to any DB is  reading which is well buffered and does not demand . OLAP vs OLTP? The first - is more than the readings, the second - is more than records. avalon/2.0.3

155

Re: Meltdown and Spectre

Hello, Evgenie Muzychenko, you wrote: I eat> Hello, lpd, you wrote: lpd>> Calls to user-mode a host all the same transit through a host kernel. I eat> That such "calls to user-mode a host" and for what it by all means to transit through its kernel? At least in kvm, the majority of virtual devices (except drivers vhost, emulated by a host kernel) are emulated user-level by the program on a host. VM at first switches in kernel-mode a host, generates event which waits user-level the emulator; after that user-level the emulator is added in queue of the scheduler and it processes request to the device when in it the context switches. How much it will be decelerated because of correction meltdown/spectre I not undertake to judge.

156

Re: Meltdown and Spectre

Hello, Evgenie Muzychenko, you wrote: Cs>> Sockets and ring  somehow together not especially live. I eat> What exactly could prevent to live it together? From outside userspace it is necessary either constant polling or the same events. And if to use events one figs of a difference already are not present - a kernel is all the same involved.

157

Re: Meltdown and Spectre

FYI https://blog.appoptics.com/visualizing-meltdown-aws/ Truth in the end write that .