1 All benchmark runs are using a fixed-size dataset over different op/s
2 values, to avoid filling the disk at high throughput.
6 Linux in-kernel NFS server, exported with the 'sync' option.
9 Same as above, but using the 'async' option so that writes need not be
10 on disk before replying to a client.
13 Initial testing with the Bluesky proxy, storing data to kvstore on a
14 local network (not to S3).
17 Testing with the Bluesky proxy writing to Amazon S3, but only using a
18 single process in the SPECsfs benchmark.
21 Standard Samba distribution set up for file sharing, tested with either
22 1 or 4 parallel processes accessing the share.