5 ms·

So we can use garbage-collected languages in trading systems without turning on the garbage collector - just collect it all at the end of the trading day. I'm j

by pradn 7y ago

So we can use garbage-collected languages in trading systems without turning on the garbage collector - just collect it all at the end of the trading day. I'm joking but this is a real technique.

magicsmoke 7y ago
It is, there was this HFT firm I interned at once that rebooted all their servers at the end of the trading day to make sure they would start clean and fast the next day.
- virtue3 7y ago
  That makes a lot of logical sense. Every single ms counts in those cases.
  - fgonzag 7y ago
    It reeks of code base problems or general misunderstanding of systems administration. Easier to just make sure you're not leaking memory (profilers exist) and actually monitoring the servers than having to reboot your servers just in case.
    - pertymcpert 7y ago
      Things like heap fragmentation still happen.
- jacques_chester 7y ago
  It also ensures you can come back up after an unexpected shutdown.
  - bob1029 7y ago
    This is the best reason. Knowing your business systems will simply come back with the power, and having that tested on a daily basis will quickly put everyone at ease.
    - gberger 7y ago
      Is this what is called chaos engineering?
- aaron_m04 7y ago
  Did they also warm up caches before open?
- xvilka 7y ago
  Now I understand why Jane Street uses OCaml. It has less of the GC problems.
fancyfish 7y ago
Yeah this is dumb but it’s actually par for the course for firms I’ve worked at. Build up all the garbage during trading hours (and on days like this hope you don’t hit resource limits) then start collecting it and rebooting at the end of the day. Most services are shut down after trading/brought up again before trading. After trading you would have accounting and other processes that would take hours. These are not done in real time and rely on daily downtime. I honestly can’t imagine these firms figuring out 24/7 trading hours.
- azernik 7y ago
  If it's dumb but it works, it isn't dumb.
bob1029 7y ago
I've never used this before, but I suspect this is exactly what you would want to use if you were on .NET and building latency-sensitive systems that see daily reboots: http://tooslowexception.com/zero-garbage-collector-for-net-core/ http://tooslowexception.com/zero-garbage-collector-for-net-c... If I was building something that was on a short fuse like this, I'd also be using structs and stack allocation as much as humanly possible before leaning onto the "having tons of physical memory" crutch. I feel like virtual memory could cover your ass for a small period of time before the whole thing started to grind to a halt.

by pradn 7y ago

So we can use garbage-collected languages in trading systems without turning on the garbage collector - just collect it all at the end of the trading day. I'm joking but this is a real technique.

magicsmoke 7y ago

It is, there was this HFT firm I interned at once that rebooted all their servers at the end of the trading day to make sure they would start clean and fast the next day.

virtue3 7y ago
That makes a lot of logical sense. Every single ms counts in those cases.
- fgonzag 7y ago
  It reeks of code base problems or general misunderstanding of systems administration. Easier to just make sure you're not leaking memory (profilers exist) and actually monitoring the servers than having to reboot your servers just in case.
  - pertymcpert 7y ago
    Things like heap fragmentation still happen.
jacques_chester 7y ago
It also ensures you can come back up after an unexpected shutdown.
- bob1029 7y ago
  This is the best reason. Knowing your business systems will simply come back with the power, and having that tested on a daily basis will quickly put everyone at ease.
  - gberger 7y ago
    Is this what is called chaos engineering?
aaron_m04 7y ago
Did they also warm up caches before open?
xvilka 7y ago
Now I understand why Jane Street uses OCaml. It has less of the GC problems.

fancyfish 7y ago

Yeah this is dumb but it’s actually par for the course for firms I’ve worked at. Build up all the garbage during trading hours (and on days like this hope you don’t hit resource limits) then start collecting it and rebooting at the end of the day. Most services are shut down after trading/brought up again before trading. After trading you would have accounting and other processes that would take hours. These are not done in real time and rely on daily downtime. I honestly can’t imagine these firms figuring out 24/7 trading hours.

azernik 7y ago
If it's dumb but it works, it isn't dumb.

bob1029 7y ago

I've never used this before, but I suspect this is exactly what you would want to use if you were on .NET and building latency-sensitive systems that see daily reboots: http://tooslowexception.com/zero-garbage-collector-for-net-core/ http://tooslowexception.com/zero-garbage-collector-for-net-c... If I was building something that was on a short fuse like this, I'd also be using structs and stack allocation as much as humanly possible before leaning onto the "having tons of physical memory" crutch. I feel like virtual memory could cover your ass for a small period of time before the whole thing started to grind to a halt.