On 12-Jul-18 18:28, The Natural Philosopher wrote:
> On 12/07/18 14:33, Alex McDonald wrote:
>> On 12-Jul-18 13:55, Ahem A Rivet's Shot wrote:
>>> On Thu, 12 Jul 2018 11:47:00 +0100
>>> The Natural Philosopher  wrote:
>>>
>>>> On 12/07/18 08:50, Ahem A Rivet's Shot wrote:
>>>>> On Thu, 12 Jul 2018 08:27:24 +0100
>>>>>     It looked that way just before NVMe, I saw some discussions
>>>>> around the possibility that DRAM may become obsolete with NVMe SSDs
>>>>> filling cache lines directly.
>>>>
>>>> Yes. That makes sense BUT it wont actually make e.g. my desktop any
>>>> faster because it spends almost no time now on IO wait
>>
>> It does spend a lot of time.
>
> Er no. Most of the time I have absolutely no reason to use any I/O at all

You said "no time on I/O wait", not that there was no to little I/O,
which is a different scenario.

>
>
> It's still worth doing a processor context
>> switch (at around 500ns) than it is waiting for the I/O, even from an
>> NVMe device; the I/O software stack is becoming a major component of
>> the delay, but it's still less than the device respnse time.
>>
>
> I fail to see where you are going with that.

In other words, I/O is very slow indeed in comaprison with the CPU, even
when using the fastest NVMe flash devices available.

>
>
>
>>
>> PM sits directly on the memory bus. In other words, you don't do block
>> IO to these new memory devices; you do loads and stores. Currently PM
>> latency is in the high hundreds of nS to single digit uS, so it sits
>> between DRAM (which is not persistent, single didgit ns to low 10s of
>> ns) and block based devices like SSDs (persistent, hundreds of µs to
>> low ms depndant on the software stack and bus).
>>
>
> So SSDS get faster. It doesnt help me because I only use the disk for
> loading programs
>
> MY bottlenecks are CPU/GPU and network.

No, SSDs don't get faster. Persistent memory is a new class of memory;
it's not based on NAND Flash, because it isn't a byte addressable
medium; that's why it's used to build disk-like systems. SSDs are
nearing their peformance barrier, and have been for a number of years.

Your bottleneck is a classic storage problem; if you're hammering
CPU/GPU *and* the network, then the data is in the wrong place.


>
>
>
>> I disagree; big data centers definitely need faster everything. The
>> more data you have and need to number crunch, the harder it becomes to
>> move it around, and the big push right now is providing very high
>> speed RDMA (memory to memory through smart network cards that don't
>> involve the CPU) type links between storage and processors to make
>> this easier.
>>
>
> It depends on the data centre.
>
> A vonsuklatnt I know works for IBM trashing thgiusanmds of PC cahssis
> runiung windoes server for comany departmets and porting the apps onto
> vurtyualised windows servers on bladess.
>
> By and large these servers exist to do very little, maybe run one
> workgroup app, and that at a really low utilisation factor. Onec in a VM
> they go into idle mode and stay there all the time
>
> Huge savings in space and power and indeed mainetance
>
> They domnt need more power, just more servers.

Sorry, even correcting the fumbled spelling, that doesn't make any sense.

>
>
>


--
Alex

--- SoupGate-Win32 v1.05