DayTrader [1] is a Websphere benchmark application
that emulates an online stock-trading system. The
application allows users to perform trading operations.
In our experiments the DB2 instance ran on the LPAR
where the crashes were injected, while the WebSphere
Application Server was running on a different LPAR.
A workload simulator was used to simulate 500 concurrent
trader clients stressing the DB2 database.
Table I summarizes the experimental results, which show
that in all cases ALDR was able to successfully resurrect
the running applications, and without much overhead. On
the average, it takes less than 18 seconds to enable ALDR,
which is a one-time cost. Our results also show that the
recovery time is independent of the type of application being
run. On the average, ALDR resurrection took only less than
3 seconds. This is 2 orders of magnitude faster than the
service downtime that Otherworld [7] have achieved for a
similar class of applications. For the vi test, the text editor
was able to recover the session as well as the buffer used
for copying/yanking data. The gzip compression test was
also successful, and gzip continued its operation after ALDR
recovery. We also tested the correctness of the compression
by de-compressing the resultant compressed file and comparing
it with the original file. The sockets applications also
continued their operations after ALDR resurrections. For
the VNC test, from a Windows VNC-clients’s viewpoint,
because the recovery time was very short, it seemed to the
user that no crash had ever happened. For the DayTrader
test, we also verified correctness by comparing the DB2
client logs generated with and without a crash, and they were
identical, which shows again that ALDR was successful in
resurrecting the DB2 connectivity with the DayTrader client