Subscribe: Comments on armstrong on software: Pure and simple transaction memories
http://armstrongonsoftware.blogspot.com/feeds/5749540209297010238/comments/default
Added By: Feedage Forager Feedage Grade B rated
Language: English
Tags:
client  database  easy  free  maybe  net  server  software  stm  sunkomutors net  sunkomutors  transaction  updates  онлайн free 
Rate this Feed
Rate this feedRate this feedRate this feedRate this feedRate this feed
Rate this feed 1 starRate this feed 2 starRate this feed 3 starRate this feed 4 starRate this feed 5 star

Comments (0)

Feed Details and Statistics Feed Statistics
Preview: Comments on armstrong on software: Pure and simple transaction memories

Comments on armstrong on software: Pure and simple transaction memories





Updated: 2017-11-14T22:35:21.332-08:00

 



[url=http://sunkomutors.net/][img]http://sunkomuto...

2009-11-25T12:36:19.960-08:00

[url=http://sunkomutors.net/][img]http://sunkomutors.net/img-add/euro2.jpg[/img][/url]
[b]where to buy photoshop in, [url=http://sunkomutors.net/]adobe air software[/url]
[url=http://sunkomutors.net/][/url] Pro 9 Advanced buy igo software
free serial number for adobe photoshop cs3 [url=http://sunkomutors.net/]microsoft software help[/url] student discount on computer software



порно подростки просмотр онлайн http://free-3x.com...

2009-11-12T15:56:39.739-08:00

порно подростки просмотр онлайн http://free-3x.com/ официальный сайт курских студентов free-3x.com/ порно молоденькие онлайн [url=http://free-3x.com/]free-3x.com[/url]



To Stefan Plantikow:It is easy to change the TM to...

2008-08-29T02:32:00.000-07:00

To Stefan Plantikow:
It is easy to change the TM to one-server-per-client mode, then what you mean by "truely parallel access" lies in the database - 'dict' here. Whatsoever, the only real difference is on which layer to do the "dirty work" - reconciling resource contentions.



Maybe my knowledge of Erlang is not deep enough, b...

2007-03-31T09:43:00.000-07:00

Maybe my knowledge of Erlang is not deep enough, but doesn't the described implementation of STM serialize all requests to the TM store? On multiprocessors this would be clearly a drawback in comparison to other STM implementations which allow truely parallel access to different variables in the same store.



Comment posting doesn't appear to work with Firefo...

2006-09-21T17:34:00.000-07:00

Comment posting doesn't appear to work with Firefox 1.5.0.7 (Javascript is enabled). I had to post this one with MSIE.



genneth wrote: A finer point is that a good STM im...

2006-09-21T17:32:00.000-07:00

genneth wrote:
A finer point is that a good STM implementation's performance (using static info to optimize certain pathways) can come to within an epsilon of a hand-crafted lock-free (or even locked) data structure, [...]

References? I don't know of any practical (even proposed) STM implementation that has an overhead for typical programs of less than a factor of 2. The



willy thanks for your poignant observation. Inaddi...

2006-09-20T11:35:00.000-07:00

willy thanks for your poignant observation.
Inaddition, STM avoid "general" concurrency related errors --deadlock, livelock,priority inversion...

What about user specific assertions...Rather aborting transaction whenever the version of any variable do not match as expected(because some other transaction interfered), one could abort if only a user given assertion over a subset of



This method is used by the Firebird database engi...

2006-09-19T18:09:00.000-07:00

This method is used by the Firebird
database engine. Every interaction with the engine occurs within a transaction. All updates etc are marked with a transaction number. There may be multiple versions of the same record. usually only the deltas between record contents are stored. This allows older transactions to see consistent snapshots of the database by only looking at record versions



I have heard this referred to as "optimistic locki...

2006-09-19T13:56:00.000-07:00

I have heard this referred to as "optimistic locking". What about starvation/priority-inversion? You have client A who updates the server frequently, and client B who updates the server very infrequently. It's easy to get into a situation where B's updates are never made to the server because B's version info never matches because A is always changing it.

You mention that this



Maybe I missed a meeting but is it not very easy t...

2006-09-19T11:18:00.000-07:00

Maybe I missed a meeting but is it not very easy to create a scenario where one client suffers starvation? Assume S, C1 and C2.

1) C1 { get, X } -> S
2) C1 <- ( X, 1, 10 ) S
3) C2 { get, X } -> S
4) C2 <- ( X, 1, 10 ) S
5) C1 { put, X, 1, 20 } -> S
6) C1 <- ( yes ) S
7) C2 { put, X, 1, 22 } -> S
8) C2 <- ( no ) S
9)



Correct me if I'm wrong, but I think STM and locks...

2006-09-19T08:18:00.000-07:00

Correct me if I'm wrong, but I think STM and locks are not excluding terms. In order to achieve STM you need either an atomic test-and-set operation in your runtime or a way to implement it... by using locks.
So when using STM you're implicitly using lock, just in a formal way where you ensure no starving nor mutual blocking nor any other typical concurrency error.



So you are saying I have to keep retrying the oper...

2006-09-19T07:42:00.000-07:00

So you are saying I have to keep retrying the operation until I get consistent sequence numbers? With a lock I can guarantee my update will succeed. There is no possibility for live lock.



I think you missed out one of the biggest gains of...

2006-09-19T02:05:00.000-07:00

I think you missed out one of the biggest gains of STM over locks: STM is composable, whereas locks aren't. This very simple property, which I'm sure any programmer can understand, translates directly into modularity and therefore simple code. A finer point is that a good STM implementation's performance (using static info to optimize certain pathways) can come to within an epsilon of a