Not an SSD, but a RAM drive.

I was thinking that with the intermittent server room issues the SETI team sees, that you wouldn't want to use a RAM drive for actual long term or citical storage, but what about using it for:

1) Preprocessing data (removing RF interference) before it's split into work units
2) Splitting work units

For all I know this might be done already in some sort of dedicated RAM drive partition in the memory of one of the servers, but if so, it's using memory that other processes might use.

A beefy non-server PC with a large RAM drive could likely process the above two operations respectably fast, and save resources for other work?

To prevent data loss, simply do not erase stored data until processing of RAM drive processed data is verified?

Some companies do make dedicated add in cards where you fill them with the amount of memory you with to use for your drive. They normally include battery modules and external power connections to prevent data loss. So you wouldn't need to get a server that supports 1TB of ram or such. You could just toss ones of these cards in each machines that needed a really fast drive to churn though data.

The big disadvantage is cost. Just having a quick look at Registered ECC dimms 8GB $150-$200, 16GB $380-$575, 32GB ~$2800. Using 16GB dimms to just make a 128GB drive would be near $3000. That kind of money can get a few enterprise class SSDs of larger size.

I have no doubt that a nice big ram drive would cut down the drive lag to nothing. I would guess the drive interconnect that is in use, FibreChannel or whatever, would get saturated long before the bandwidth of the RAM drive was fully utilized.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Some companies do make dedicated add in cards where you fill them with the amount of memory you with to use for your drive. They normally include battery modules and external power connections to prevent data loss. So you wouldn't need to get a server that supports 1TB of ram or such. You could just toss ones of these cards in each machines that needed a really fast drive to churn though data.

The big disadvantage is cost. Just having a quick look at Registered ECC dimms 8GB $150-$200, 16GB $380-$575, 32GB ~$2800. Using 16GB dimms to just make a 128GB drive would be near $3000. That kind of money can get a few enterprise class SSDs of larger size.

I have no doubt that a nice big ram drive would cut down the drive lag to nothing. I would guess the drive interconnect that is in use, FibreChannel or whatever, would get saturated long before the bandwidth of the RAM drive was fully utilized.

That is the truth. I stumbled across the HyperDrive5 a year or two ago and drooled about the possibilities that it could deliver. HD5 has an onboard ECC chip, so you can get the cheapest RAM you can find and throw it in there. Problem is that it only goes to 64gb (8x8gb), and it is on sata 3gb/s, but it does have two SATA ports (one for each 4x bank), so you could run it in raid-0 on two channels.

From the benchmarks that I've seen with it, there are SSDs on the market right now that are pretty close (~75%) to the same stats for throughput and IOPS. And they're about 75% less in cost. The cost of doing an HD5 with 64gb of memory: 500 for the device itself, and then whatever 8x8gb of DDR2 is. If you want to utilize the HD5 to its fullest capacity, you also need a CompactFlash card to use as an emergency backup for it.

In all honesty.. a good SSD is the best way to go. More capacity at a fraction of the cost, and you can just about max out the SATA-II 3gb/s interface with them as they are right now. Some models have already moved over to the 6gb/sec interface to surpass that limit.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

I'm wondering if this thread be merged with the one I started a few weeks ago.