From: Brian Warner <warner@lothar.com>
Date: Fri, 7 Sep 2007 06:01:07 +0000 (-0700)
Subject: provisioning: explain the 'survival' number better
X-Git-Url: https://git.rkrishnan.org/architecture.txt?a=commitdiff_plain;h=5561d3afe4011e94d40143d41bc8b025f8272ba3;p=tahoe-lafs%2Ftahoe-lafs.git

provisioning: explain the 'survival' number better
---

diff --git a/src/allmydata/provisioning.py b/src/allmydata/provisioning.py
index b07c6b8e..ae0cd8a7 100644
--- a/src/allmydata/provisioning.py
+++ b/src/allmydata/provisioning.py
@@ -597,6 +597,13 @@ class ProvisioningTool(rend.Page):
                        T.div["share data loss rate: ",
                              number(share_data_loss_rate,"Bps")])
 
+            # the worst-case survival numbers occur when we do a file check
+            # and the file is just above the threshold for repair (so we
+            # decide to not repair it). The question is then: what is the
+            # chance that the file will decay so badly before the next check
+            # that we can't recover it? The resulting probability is per
+            # check interval.
+            # Note that the chances of us getting into this situation are low.
             P_disk_failure_during_interval = (drive_failure_rate *
                                               file_check_interval)
             disk_failure_dBF = 10*math.log10(P_disk_failure_during_interval)
@@ -614,6 +621,7 @@ class ProvisioningTool(rend.Page):
                              user_files_survives_dBA,
                              "all files in grid = %d dBA" %
                              all_files_survives_dBA,
+                             " (per worst-case check interval)",
                              ])