Low latency framework: the must read presentation of Richard Croucher
I've just read a very informative document written by Richard Croucher about how to build low latency systems. One of the best paper I've ever read on my favorite topic (very synthetic and clear).
Very interesting, but some details bothers me : Slide 11 "Stretched VLAN can be used to remove router insertion latency cost"
This is not really true as when using extended VLANs, L2 packets are encapsulated in L3 ones. So there will always be routers on the network path, it's transparent, but it's not direct switch to switch communication.
Slide 21 I heard that noop scheduling isn't the best. To improve latency, the deadline scheduler is recommended, as it's based on service time concept rather than FIFO queue. Good article to read ; http://www.redhat.com/magazine/008jun05/features/schedulers/
Anyway, it's really hard to identify I/O workload on systems, so there's no perfect scheduler because processes are not behaving the same way. Would be great to have OS I/O and Applicative I/O all seggregated (schedulers, interruptions, storage) :)
Very interesting, but some details bothers me :
ReplyDeleteSlide 11
"Stretched VLAN can be used to remove router insertion latency cost"
This is not really true as when using extended VLANs, L2 packets are encapsulated in L3 ones.
So there will always be routers on the network path, it's transparent, but it's not direct switch to switch communication.
Slide 21
I heard that noop scheduling isn't the best.
To improve latency, the deadline scheduler is recommended, as it's based on service time concept rather than FIFO queue.
Good article to read ; http://www.redhat.com/magazine/008jun05/features/schedulers/
Anyway, it's really hard to identify I/O workload on systems, so there's no perfect scheduler because processes are not behaving the same way. Would be great to have OS I/O and Applicative I/O all seggregated (schedulers, interruptions, storage) :)