skyde

53 ms·

571.
by skyde 8y ago
How does this compare to Adaptive Radix Tree (ART) with optimistic lock coupling? Or is it more about transactions isolation than thread safe access ?
572.
by skyde 8y ago
very interesting could you please give more detail on how Tandem and Space Shuttle Flight Computer relied on software support ?
573.
by skyde 8y ago
"your app had to be either a) stateless or b) cluster-aware" could you elaborate on what your app had to do to be cluster-aware ?
574.
by skyde 8y ago
I could not find anywhere an explanation of what HP nonstop is actually doing. The fact you say its like VMWare make me think you mean VMware vLockstep. Documentation say : "While both the primary and secondary VMs receive the same inp
575.
by skyde 8y ago
not sure what you mean by "on the wire". But even if this is assumed to be the time to transfer a UDP packet from one network card buffer to another network card buffer directly connected by a cable this seem extremely low.
576.
by skyde 8y ago
can someone explain this part to me Send 2000 byte over network = 44ns Roundtrip in same datacenter = 500000 ns Isn't a rountrip simply a send from A to B and then a send from node B to node A?
577.
by skyde 8y ago
I had to opportunity to look at the code. I believe this should be compared against embedded hash database such as "kyoto cabinet, LevelDB, RocksDB" . They introduce a novel latch free hashtable that they say is faster than other
578.
by skyde 8y ago
On android this Is done using LRU cache for background app. If your app has a cached process and it retains memory that it currently does not need, then your app—even while the user is not using it— affects the system's overall perform
579.
by skyde 9y ago
It’s not clear from the paper how the rank is updated. If anyone understand better, could you explain. The eviction is doing random sampling of N entry and is straight forward. But how does it compute the hit density online per entry.
580.
by skyde 9y ago
:(
581.
by skyde 9y ago
This is very sad, I really hope open Solaris can keep alive without Oracle.
582.
by skyde 9y ago
Haskell is doing this already with par and seq function to control parallel and lazy evaluation. I believe we should have something like this for Async
583.
by skyde 9y ago
Where can I get this :)
584.
by skyde 9y ago
No I was talking exactly about Predicate Locking. They detect writes which conflict with earlier reads I dont think this could be called 2PL because it's only used to detect conflict. Correctness requires that any insert into an index
585.
by skyde 9y ago
I think what your looking for is called range lock. When a query sort or filter rows using a range predicate. If using repeatable read isolation. A range lock is created to detect conflict with another transaction inserting new matching row
586.
by skyde 9y ago
So the main problem seem to be that libaio on linux does not use the filesystem cache because it's forcing the use of O_Direct. Any reason linux could not support AIO on regular file like FreeBSD ?
587.
by skyde 9y ago
I woulds strongly advise against using Redis for lock. Maybe something with a real consensus algorithm like ETCD or Zookeeper but otherwise it's better to keep the lock inside the RDBMS.
588.
by skyde 9y ago
Could you elaborate about what you don't like about it
589.
by skyde 9y ago
Isn't that what BFQ scheduler was designed for ?
590.
by skyde 9y ago
While Network socket in Go are using epolls automatically, file are not. From looking at Badger code for example: fd.ReadAt(buf, offset) would block. See this issue: https://github.com/golang/go/issues/6817
591.
by skyde 9y ago
Actually you can still put nginx inside a cgroup and have nginx read the loadAvg inside the CGROUP instead of the global host loadAVG by calling getloadavg();
592.
by skyde 9y ago
To fill the queue on Linux goroutine wont be enough you would need to use libaio directly. sudo apt-get install libaio1 libaio-dev.
593.
by skyde 9y ago
This is amazing!!
594.
by skyde 9y ago
Range iteration latency is very important and might be limited by concurrency. I think you can only get 100K IOPS on Amazon’s i3.large when the disk Request queue is full. fio [1] can easily do this because it spawn a number of threads Whil
595.
by skyde 9y ago
Would adding this only to the http reverse proxy machines provide most of the benefit without have to patch all servers. This seem to have the greatest effect over wan links.
596.
by skyde 9y ago
How can we use it today! Is it in Linux code already and easy to enable ?
597.
by skyde 10y ago
100% agree, if you can afford to use GRPC. It's much simpler than having to use swagger.
598.
by skyde 10y ago
Thanks a lot for clarification.
599.
by skyde 10y ago
Could you explain how Bizure compare to Gryadka. Are the performance tradeoff similar.
600.
by skyde 10y ago
it seem that the performance gain of not having all update go to a single log could be also achieved by having one log per bucket. My understanding is that with Bizur all update for the same bucket also need to proceed sequentially. Also wi

More ›