4 ms·
I'm all for options, but isn't swap slow enough as it is already without adding additional overhead? Even on tiny embedded devices, storage is cheaper than cyc
by TheCraiggers 13y ago
I'm all for options, but isn't swap slow enough as it is already without adding additional overhead?
Even on tiny embedded devices, storage is cheaper than cycles IMO.
- s_baby 13y agoI think they're doing compressed swap to ramdisk.
- nly 13y agoIt's not the cost of the storage, it's how slow it to read and write to it. This is still going to be an order of magnitude faster than swapping to disk.
- bodyfour 13y agoAs others have pointed out, the point of zswap is to "swap" in and out of RAM. I'd like to add that your premise is faulty anyway: since compressed objects are smaller doing I/O on them is faster. As long as cost of the I/O you saved is less than the cost of doing the compression, it's faster not slower. Transparently compressed I/O usually uses high-speed algorithms (LZ4, LZO, etc) to take advantage of this effect. Also, when a machine is swapping you'll usually find that the CPU is near idle since useful work is blocked on the swap I/O. Spending a little CPU time to speed up that I/O is a win.
- TheCraiggers 13y agoAn enlightening response. Thank you.
- aaronbee 13y agoThe point is to avoid writing to disk entirely by compressing RAM that may otherwise be swapped to disk when remaining RAM is low. See http://thread.gmane.org/gmane.linux.kernel.mm/91969 http://thread.gmane.org/gmane.linux.kernel.mm/91969