RAMdisk

[suchamoneypit]

So I've had an idea. I can fit another 128GB of RAM in my server with my current config. I have been wanting to buy a very high speed PCI based SSD (2+GB/s read) , a 120GB size. I would then make this 128GB of extra RAM a RAMdisk and copy my entire PCI SSD (that has all my servers) onto the RAMdisk, and then have my 74GB of current RAM left over for actually running the servers. Theoretically, when I boot my PC, everything that runs would be very quickly transferred straight into a RAMdisk. As far as storage goes you couldn't go much faster than that.

 

Is it practical? If it worked it certainly would be awesome if only for being super cool and unique.

[Electronics Wizardy]

Not practical, but you can use the ram as a cache. What os and hypervisor are you running?

[laufersteppenwolf]

in theory, you could definitely make it work.

However, you have to keep in mind that in case of any critical errors causing a reboot occurs, all your data would be gone. You'd have to run ongoing rsync tasks in order to keep your SSD up to date with the data stored on your RAM

 

But as already suggested, using it as cache would definitely be easier and more practical

[suchamoneypit]

15 minutes ago, Electronics Wizardy said:

Not practical, but you can use the ram as a cache. What os and hypervisor are you running?

I am using Windows server 2012 R2 ( I understand this is easier to do on Linux)

 

[leadeater]

1 hour ago, suchamoneypit said:

I am using Windows server 2012 R2 ( I understand this is easier to do on Linux)

 

Can't really see why you would need to, PCI-E NVMe SSD's are very close to RAM speeds as it is so you are unlikely to see any real performance improvement between the two. Then there is the whole issue of the VMs actually being able to use that speed, more VMs the better utilization wise.

 

You can configure tiered storage spaces and use the RAM disk for that but storage spaces assumes that the SSD tier is persistent storage so would corrupt the pool on reboot. Pure speed wise two small PCI-E NVMe disks in a tiered storage pool would perform as fast as your original plan would.

[suchamoneypit]

1 hour ago, leadeater said:

Can't really see why you would need to, PCI-E NVMe SSD's are very close to RAM speeds as it is so you are unlikely to see any real performance improvement between the two. Then there is the whole issue of the VMs actually being able to use that speed, more VMs the better utilization wise.

 

You can configure tiered storage spaces and use the RAM disk for that but storage spaces assumes that the SSD tier is persistent storage so would corrupt the pool on reboot. Pure speed wise two small PCI-E NVMe disks in a tiered storage pool would perform as fast as your original plan would.

hmm, using two PCI-E NVMe in a raid could be a better option. 1 PCI-SSD +128GB RAM is $400, but getting two PCI SSDs is also $400. Just running 2 of the PCI-SSDs in a raid for speed might be an easier option (I would definitely want to backup data to an HDD though in case of a raid failure).

[leadeater]

1 hour ago, suchamoneypit said:

hmm, using two PCI-E NVMe in a raid could be a better option. 1 PCI-SSD +128GB RAM is $400, but getting two PCI SSDs is also $400. Just running 2 of the PCI-SSDs in a raid for speed might be an easier option (I would definitely want to backup data to an HDD though in case of a raid failure).

Since you are using Server 2012 R2 have a look at Storage Spaces in stead of RAID, the SSDs can be used to cache the HDDs and I can say it works extremely well. This way you can set the SSD tier to 2 way mirror and get the performance of both for reads and one for writes with the benefit of redundancy.

 

Here's a pool of 6 Samsung Pro's and 5 3TB Reds, PCI-E NVMe would significantly improve the small block speeds.