Starting today, hard-disk servers set their own read limits, at the point where their disks get the most reading done. On three servers where the first limits were tighter than the disks needed, this let 1.6 to 2.4 times more data be read per second.
Yesterday's per-account read limits (announcement 725) made small reads on shared disks answer much faster. On some hard-disk servers, though, the first settings were tighter than the disks needed. The disks there answered a read in about 13 ms, while a busy hard-disk array does its best work at around 20-25 ms, so they had room to spare. On three of them, setting the limits five times looser gave that 1.6 to 2.4 times more reading, with less time stuck waiting for reads.
What changed
Every hour, each hard-disk server switches between two settings, one twice as loose as the other, and counts the reads every account actually completed. If the looser setting delivers more than 10% more, the server loosens. If it does not, the server tightens, because that extra room would only let heavy readers crowd the queue again. One step takes about twelve hours of real traffic, and the server keeps re-testing as its load changes. Your share of the limit still follows your disk priority, as before.
Why it is tuned live
No single number fits every server. The best setting for a disk depends on how many accounts read at once, how large their reads are and what else the array is doing. Only production load shows how a setting scales and which one is right. So each server now measures its own, instead of trusting one number from me.
Is it on your server?
It comes with the platform update, and the first hard-disk servers got it today. There is nothing to switch on. Your info tab still shows "Disk reads for your account right now: up to N per second". On a hard-disk server that number can change from hour to hour while the server tests its settings.
Best Regards,
Väinämöinen
Pulsed Media Support
(Strung the kantele too tight; retuned by ear)
Saturday, November 7, 2026
