]> git.rkrishnan.org Git - tahoe-lafs/tahoe-lafs.git/commitdiff
tool to generate an overhead/alacrity table for various hashing schemes
authorBrian Warner <warner@lothar.com>
Sun, 22 Oct 2006 03:10:04 +0000 (20:10 -0700)
committerBrian Warner <warner@lothar.com>
Sun, 22 Oct 2006 03:10:04 +0000 (20:10 -0700)
sizes.py [new file with mode: 0644]

diff --git a/sizes.py b/sizes.py
new file mode 100644 (file)
index 0000000..cf19b07
--- /dev/null
+++ b/sizes.py
@@ -0,0 +1,204 @@
+#! /usr/bin/python
+
+import random, math, os, re
+from twisted.python import usage
+
+class Args(usage.Options):
+    optParameters = [
+        ["mode", "m", "alpha", "validation scheme"],
+        ["arity", "k", 2, "k (airty) for hash tree"],
+        ]
+    def opt_arity(self, option):
+        self['arity'] = int(option)
+    def parseArgs(self, *args):
+        if len(args) > 0:
+            self['mode'] = args[0]
+
+
+def charttest():
+    import gdchart
+    sizes = [random.randrange(10, 20) for i in range(10)]
+    x = gdchart.Line()
+    x.width = 250
+    x.height = 250
+    x.xtitle = "sample"
+    x.ytitle = "size"
+    x.title = "Example Graph"
+    #x.ext_color = [ "white", "yellow", "red", "blue", "green"]
+    x.setData(sizes)
+    #x.setLabels(["Mon", "Tue", "Wed", "Thu", "Fri"])
+    x.draw("simple.png")
+
+KiB=1024
+MiB=1024*KiB
+GiB=1024*MiB
+TiB=1024*GiB
+
+class Sizes:
+    def __init__(self, mode, file_size, arity=2):
+        MAX_SEGSIZE = 1*MiB
+        self.mode = mode
+        self.file_size = file_size
+        self.seg_size = seg_size = 1.0 * min(MAX_SEGSIZE, file_size)
+        self.num_segs = num_segs = math.ceil(file_size / seg_size)
+        self.num_subblocks = num_subblocks = num_segs
+
+        self.num_blocks = num_blocks = 100
+        self.blocks_needed = blocks_needed = 25
+
+        self.subblock_size = subblock_size = seg_size / blocks_needed
+        self.block_size = block_size = subblock_size * num_subblocks
+
+        # none of this includes the block-level hash chain yet, since that is
+        # only a function of the number of blocks. All overhead numbers
+        # assume that the block-level hash chain has already been sent,
+        # including the root of the subblock-level hash tree.
+
+        if mode == "alpha":
+            # no hash tree at all
+            self.subblock_arity = 0
+            self.subblock_tree_depth = 0
+            self.subblock_overhead = 0
+            self.bytes_until_some_data = 20 + block_size
+            self.block_storage_overhead = 0
+            self.block_transmission_overhead = 0
+
+        elif mode == "beta":
+            # k=num_subblocks, d=1
+            # each subblock has a 20-byte hash
+            self.subblock_arity = num_subblocks
+            self.subblock_tree_depth = 1
+            self.subblock_overhead = 20
+            # the block has a list of hashes, one for each subblock
+            self.block_storage_overhead = (self.subblock_overhead *
+                                           num_subblocks)
+            # we can get away with not sending the hash of the block that
+            # we're sending in full, once
+            self.block_transmission_overhead = self.block_storage_overhead - 20
+            # we must get the whole list (so it can be validated) before
+            # any data can be validated
+            self.bytes_until_some_data = (self.block_transmission_overhead +
+                                          subblock_size)
+
+        elif mode == "gamma":
+            self.subblock_arity = k = arity
+            d = math.ceil(math.log(num_subblocks, k))
+            self.subblock_tree_depth = d
+            num_leaves = k ** d
+            # to make things easier, we make the pessimistic assumption that
+            # we have to store hashes for all the empty places in the tree
+            # (when the number of blocks is not an exact exponent of k)
+            self.subblock_overhead = 20
+            # the subblock hashes are organized into a k-ary tree, which
+            # means storing (and eventually transmitting) more hashes. This
+            # count includes all the low-level block hashes and the root.
+            hash_nodes = (num_leaves*k - 1) / (k - 1)
+            #print "hash_depth", d
+            #print "num_leaves", num_leaves
+            #print "hash_nodes", hash_nodes
+            # the storage overhead is this
+            self.block_storage_overhead = 20 * (hash_nodes - 1)
+            # the transmission overhead is smaller: if we actually transmit
+            # every subblock, we don't have to transmit 1/k of the
+            # lowest-level subblock hashes, and we don't have to transmit the
+            # root because it was already sent with the block-level hash tree
+            self.block_transmission_overhead = 20 * (hash_nodes
+                                                     - 1 # the root
+                                                     - num_leaves / k)
+            # we must get a full sibling hash chain before we can validate
+            # any data
+            sibling_length = d * (k-1)
+            self.bytes_until_some_data = 20 * sibling_length + subblock_size
+            
+            
+
+        else:
+            raise RuntimeError("unknown mode '%s" % mode)
+
+        self.storage_overhead = self.block_storage_overhead * num_blocks
+        self.storage_overhead_percentage = 100.0 * self.storage_overhead / file_size
+
+    def dump(self):
+        for k in ("mode", "file_size", "seg_size",
+                  "num_segs", "num_subblocks", "num_blocks", "blocks_needed",
+                  "subblock_size", "block_size",
+                  "subblock_arity", "subblock_tree_depth",
+                  "subblock_overhead",
+                  "block_storage_overhead", "block_transmission_overhead",
+                  "storage_overhead", "storage_overhead_percentage",
+                  "bytes_until_some_data"):
+            print k, getattr(self, k)
+
+def fmt(num, trim=False):
+    if num < KiB:
+        s = str(num) + "#"
+    elif num < MiB:
+        s = "%.2fk" % (num / KiB)
+    elif num < GiB:
+        s = "%.2fM" % (num / MiB)
+    elif num < TiB:
+        s = "%.2fG" % (num / GiB)
+    else:
+        s = "big"
+    if trim:
+        s = re.sub(r'(\.0+)([kMG#])',
+                   lambda m: m.group(2),
+                   s)
+    else:
+        s = re.sub(r'(\.0+)([kMG#])',
+                   lambda m: (" "*len(m.group(1))+m.group(2)),
+                   s)
+    if s.endswith("#"):
+        s = s[:-1] + " "
+    return s
+
+def text():
+    opts = Args()
+    opts.parseOptions()
+    mode = opts["mode"]
+    arity = opts["arity"]
+    #      0123456789012345678901234567890123456789012345678901234567890123456
+    print "mode=%s" % mode, " arity=%d" % arity
+    print "                    storage    storage"
+    print "Size     blocksize  overhead   overhead     k  d  alacrity"
+    print "                    (bytes)      (%)"
+    print "-------  -------    --------   --------  ---- --  --------"
+    sizes = [2 ** i for i in range(7, 32)]
+    for file_size in sizes:
+        s = Sizes(mode, file_size, arity)
+        out = ""
+        out += "%7s  " % fmt(file_size, trim=True)
+        out += "%7s    " % fmt(s.block_size)
+        out += "%8s" % fmt(s.storage_overhead)
+        out += "%10.2f  " % s.storage_overhead_percentage
+        out += " %4d" % int(s.subblock_arity)
+        out += " %2d" % int(s.subblock_tree_depth)
+        out += " %8s" % fmt(s.bytes_until_some_data)
+        print out
+
+
+def graph():
+    # doesn't work yet
+    import Gnuplot
+    opts = Args()
+    opts.parseOptions()
+    mode = opts["mode"]
+    arity = opts["arity"]
+    g = Gnuplot.Gnuplot(debug=1)
+    g.title("overhead / alacrity tradeoffs")
+    g.xlabel("file size")
+    g.ylabel("stuff")
+    sizes = [2 ** i for i in range(7, 32)]
+    series = {"overhead": {}, "alacrity": {}}
+    for file_size in sizes:
+        s = Sizes(mode, file_size, arity)
+        series["overhead"][file_size] = s.storage_overhead_percentage
+        series["alacrity"][file_size] = s.bytes_until_some_data
+    g.plot([ (fs, series["overhead"][fs])
+             for fs in sizes ])
+    raw_input("press return")
+
+
+if __name__ == '__main__':
+    text()
+    #graph()