Cache Misses, and Why the Fast Version Was Slow
#performance#ops
the query that was fine until it wasn't
#performance#ops
the fast version was slower, and the cache was why
#performance#ops
what a syscall actually costs, with numbers
#performance#ops
perf top on a box that had no business being busy
#performance#ops
what a syscall actually costs you, with numbers
#performance#ops
how much does crossing into the kernel actually cost
#performance#ops
the average is lying to you, look at p99
#performance#ops
watching syscalls without a debugger holding the door open
#performance#ops
the query that wasn't slow, it was just run a million times
#performance#ops
the fast version was slower, and the cache told me why
#performance#ops
the average response time that hid a fire
#performance#ops
the query that was quietly killing us
#performance#ops
the cost of a syscall, measured
#performance#ops
the query that was fine until it wasn't
#performance#ops
the fast version was slower, and the cache told me why
#performance#ops
cache misses, and why the fast version was slower
#performance#ops
perf top on a box that should have been idle
#performance#ops
the flamegraph that pointed at the wrong hero
#performance#ops
the query that wasn't slow, it was just always running
#performance#ops
a database query that was quietly killing us
#performance#ops
perf top on a box that shouldn't be busy
#performance#ops
the query that was fine until it wasn't
#performance#ops
the average latency was fine, which is why everyone was angry
#performance#ops
the box that was busy doing nothing
#performance#ops
the idle box that wasn't
#performance#ops
the flame graph that pointed at a function i'd forgotten existed
#performance#ops
perf top, and an idle box that wasn't
#performance#ops
what a syscall actually costs you
#performance#ops
the flamegraph that pointed at logging, of all things
#performance#ops
how much does a syscall actually cost?
#performance#ops
flamegraphs and a hot path i never suspected
#performance#ops
the cost of a syscall, measured
#performance#ops
the average latency that lied to my face
#performance#ops
the flamegraph that pointed at the wrong file entirely
#performance#ops
the optimisation that made everything slower
#performance#ops
the flamegraph pointed somewhere stupid
#performance#ops
the average was fine and the customers were furious
#performance#ops
the query that was fine until it wasn't
#performance#ops
the optimisation that made everything slower
#performance#ops
the average latency is lying to you
#performance#ops
the query that only got slow once it mattered
#performance#ops
the cost of a syscall, measured
#performance#ops
ebpf, or how i stopped guessing and watched the kernel
#performance#ops
latency, p99, and the averages that lie to you
#performance#ops
ebpf, or finally being able to ask the kernel a question
#performance#ops
the flamegraph pointed at the one function i'd ruled out
#performance#ops
the rewrite that was meant to be faster, and was not
#performance#ops
the query that was quietly killing us
#performance#ops
the fast version was slower, and the cache told me why
#performance#ops
bpftrace, and finally being able to ask the kernel a question
#performance#ops
the average is fine, the p99 is on fire
#performance#ops
the query that was fine until it wasn't
#performance#ops
the idle box that was pegging a core
#performance#ops
perf top on a box that should have been idle
#performance#ops
the flamegraph that pointed at the one function i trusted
#performance#ops
an idle box at forty percent cpu, and what perf top told me
#performance#ops
what does a syscall actually cost?
#performance#ops
watching syscalls without a debugger in sight
#performance#ops
the query that ate a node every afternoon
#performance#ops
the "optimised" loop that ran slower than the naive one
#performance#ops
the average is lying to you
#performance#ops
the query that was bleeding us for months
#performance#ops
perf top on a box that shouldn't be busy
#performance#ops
perf top on a box that shouldn't be busy
#performance#ops
watching the kernel work, finally, with ebpf
#performance#ops
perf top on a box that had no business being busy
#performance#ops
a database query that was quietly killing us
#performance#ops
the query that worked fine until it didn't
#performance#ops
the average is fine, which is exactly the problem
#performance#ops
who is eating all the cpu on an idle server
#performance#ops
the query that wasn't slow, just slow ten thousand times
#performance#ops
the orm that hid a thousand queries
#performance#ops
how much does a syscall actually cost
#performance#ops
the idle server that wasn't idle
#performance#ops
the flamegraph that pointed at the wrong thing, then the right one
#performance#ops
the day i could finally watch the kernel work
#performance#ops
the average that hid the outage
#performance#ops
watching the kernel without a debugger
#performance#ops
the average is lying to you, look at p99
#performance#ops
eBPF, or finally being able to ask the kernel what it's doing
#performance#ops
why the average latency was lying to me
#performance#ops
perf top on a box that should have been asleep
#performance#ops
ebpf, and finally seeing what the kernel sees
#performance#ops
perf top on a box that was meant to be idle
#performance#ops
eBPF, or finally seeing what the kernel sees
#performance#ops
the flamegraph that pointed at the wrong thing, which was the right thing
#performance#ops
ebpf, and finally seeing what the kernel sees
#performance#ops
what a syscall actually costs you
#performance#ops
the idle server that was burning a whole core
#performance#ops
the query that was bleeding us a hundred milliseconds at a time
#performance#ops
the flamegraph that pointed at the one function i'd never have profiled
#performance#ops
the flamegraph that pointed at the last function I'd have guessed
#performance#ops
the N+1 that hid behind a fast endpoint
#performance#ops
the join that got expensive when nobody was looking
#performance#ops
the hot path was in the logging
#performance#ops
how much does crossing into the kernel actually cost
#performance#ops
ebpf, and finally seeing what the kernel sees
#performance#ops
ebpf, or how i stopped guessing about a slow box
#performance#ops
instrumenting the kernel without rebooting it
#performance#ops
the flamegraph that pointed at the one line i'd never have guessed
#performance#ops
what a syscall actually costs, with numbers
#performance#ops
your average latency is lying to you
#performance#ops
the average latency is lying to you
#performance#ops
what a syscall actually costs you
#performance#ops
your average latency is lying to you
#performance#ops
the query that was quietly killing us
#performance#ops
a syscall is about 80 nanoseconds, so stop making so many
#performance#ops
the flamegraph found the thing I'd sworn was fine
#performance#ops
how much does a syscall actually cost?
#performance#ops
the flamegraph that pointed at the one function i'd never have guessed
#performance#ops
the hot path was in the logging
#performance#ops
flamegraphs and a hot path i never suspected
#performance#ops
the flamegraph that pointed at the wrong file entirely
#performance#ops