4 ms·

Are cloud based memory architectures the next big thing?

20 points by helwr 16y ago

Khroma 16y ago
So basically.. cloud based caching?
zitterbewegung 16y ago
They may be the next big thing since the old guard databases although are optimized for disk in memory eliminate that due to decreased latency. The caveat is that the in memory databases aren't as mature as databases that persist to disk.
jedberg 16y ago
># Performance is better because data is accessed from memory instead of through a database to a disk. ># Scalability is linear because as more servers are added data is transparently load balanced across the servers so there is an automated in-memory sharding. ># Availability is higher because multiple copies of data are kept in memory and the entire system reroutes on failure. ># Application development is faster because there’s only one layer of software to deal with, the cache, and its API is simple. All the complexity is hidden from the programmer which means all a developer has to do is get >and put data. All but the first thing apply to disk based databases too... At least the new ones like Cassandra, as well as the old ones with the right packages.
- gxti 16y ago
  Buy enough RAM to fit the entire dataset in memory and any database will scream.
jasonkester 16y ago
I've been a fan of this sort of thing for several years now. Of course back in the day it would get you complained at for violating the best-practice of not keeping heavy objects in the Application or Session state (presumably because back then you only had 500mb of it). If you go play around on Twiddla today, chances are nothing you do there will end up on physical storage, ever. All the sandboxes and guest sessions spend their entire lives in memory, and only in very rare cases get persisted to the database these days. One side effect of doing this is that it makes scaling out a lot easier. If we ever hit a traffic spike that bothers our server, I can spin up a few EC2 instances and hand off the actual Twiddling to them without having to worry about what to do about getting the data written back to the database. It also makes a nice differentiator between free and paid plans: If you need the ability to save your stuff, we're happy to let you pay to do so. If not, go nuts, since it doesn't cost us anything to keep you in memory.
jermy 16y ago
Odd that SSDs aren't mentioned in the article at all - they would mostly avoid many of the criticisms levelled at disks without requiring redevelopment
ekidd 16y ago
That's an interesting article, although the level of hype for cloud computing is rather high (and I'm an enormous fan of cloud computing): "When PCs and client/server computing came along, the departmental computing guys (i.e. DEC), considered PCs to be a special case of smart terminals... So the batchguys are dead, the timesharing guys are dead, the departmental computing guys are dead, and the client server guys are dead. Notice a pattern?" That said, there are already several hybrid memory/disk storage layers that might actually be a better choice than the pure in-memory approach that the author suggests. For example, MongoDB is basically an in-memory document database that's mmapped to a disk file and flushed once per minute. CouchDB, IIRC, is basically an in-memory database backed by an append-only transaction log on disk. These sorts of designs should give you many of the bandwidth advantages of pure memory systems, but still maintain the ability to survive a power outage.
0bfusct3 16y ago
Can we just call it what it's been known as in computer science since oh 1960's distributed operating systems?

20 points by helwr 16y ago

So basically.. cloud based caching?

They may be the next big thing since the old guard databases although are optimized for disk in memory eliminate that due to decreased latency. The caveat is that the in memory databases aren't as mature as databases that persist to disk.

jedberg 16y ago

># Performance is better because data is accessed from memory instead of through a database to a disk. ># Scalability is linear because as more servers are added data is transparently load balanced across the servers so there is an automated in-memory sharding. ># Availability is higher because multiple copies of data are kept in memory and the entire system reroutes on failure. ># Application development is faster because there’s only one layer of software to deal with, the cache, and its API is simple. All the complexity is hidden from the programmer which means all a developer has to do is get >and put data. All but the first thing apply to disk based databases too... At least the new ones like Cassandra, as well as the old ones with the right packages.

gxti 16y ago
Buy enough RAM to fit the entire dataset in memory and any database will scream.

jasonkester 16y ago

I've been a fan of this sort of thing for several years now. Of course back in the day it would get you complained at for violating the best-practice of not keeping heavy objects in the Application or Session state (presumably because back then you only had 500mb of it). If you go play around on Twiddla today, chances are nothing you do there will end up on physical storage, ever. All the sandboxes and guest sessions spend their entire lives in memory, and only in very rare cases get persisted to the database these days. One side effect of doing this is that it makes scaling out a lot easier. If we ever hit a traffic spike that bothers our server, I can spin up a few EC2 instances and hand off the actual Twiddling to them without having to worry about what to do about getting the data written back to the database. It also makes a nice differentiator between free and paid plans: If you need the ability to save your stuff, we're happy to let you pay to do so. If not, go nuts, since it doesn't cost us anything to keep you in memory.

jermy 16y ago

Odd that SSDs aren't mentioned in the article at all - they would mostly avoid many of the criticisms levelled at disks without requiring redevelopment

ekidd 16y ago

That's an interesting article, although the level of hype for cloud computing is rather high (and I'm an enormous fan of cloud computing): "When PCs and client/server computing came along, the departmental computing guys (i.e. DEC), considered PCs to be a special case of smart terminals... So the batchguys are dead, the timesharing guys are dead, the departmental computing guys are dead, and the client server guys are dead. Notice a pattern?" That said, there are already several hybrid memory/disk storage layers that might actually be a better choice than the pure in-memory approach that the author suggests. For example, MongoDB is basically an in-memory document database that's mmapped to a disk file and flushed once per minute. CouchDB, IIRC, is basically an in-memory database backed by an append-only transaction log on disk. These sorts of designs should give you many of the bandwidth advantages of pure memory systems, but still maintain the ability to survive a power outage.

0bfusct3 16y ago

Can we just call it what it's been known as in computer science since oh 1960's distributed operating systems?