Marek Behr from the Jülich-Aachen Research Alliance uses computer simula-tions to develop life-saving blood pumps.
then has a toxic effect on the patient and the damaged blood cells can no longer transport any oxygen. Nor should the blood platelets be damaged or activated if at all possible otherwise blood clots may be easily formed which then block the blood ves
sels resulting in a lifethreatening throm
bosis. However, trial and error cannot be used on patients in order to improve the pump in such a way that it does less dam
age to the blood it transports. This is where calculations of flow conditions on the computer can come into play. “This is a typical case where simulations can be used. They help when experiments are
impossible, too resourceconsuming or too dangerous,” explains Prof. Felix Wolf from the Jülich Supercomputing Centre.
“In the first phase up to 2001, we developed the simulation to such an extent that we can be sure that it corre
sponds to the results of real experi
ments.” This is how Behr describes the early stages of the project. In the second phase of development, they succeeded in revealing where the blood was damaged most while flowing through the pump.
“This occurred to a greater extent than previously assumed in the rear section of the pump, in what is known as the dif
fuser,” Behr explains. These findings have already influenced the design of a new version of the DeBakey pump. Behr con
tinues: “Since 2006 in the third stage of the project, we have been developing mathematical methods and programs in order to optimize the pump’s mode of operation.”
Blood also makes special demands on the simulation. Whereas the flow be
haviour of water can be calculated on a normal PC, a highperformance super
computer such as Jülich’s JUGENE is required to investigate the liquid flow in such a complicated pump and to study the dynamics of the mixture of liquid and One result of the simulation: the regions of the blood pump shown in red experience the
highest pressure during operation, and those in blue the lowest.
small so that it can be implanted into the human body,” explains Behr. “How
ever, in order to provide sufficient sup
port for the heart it must be able to pump several litres of blood in a minute. Strong shearing forces are the result when so much liquid flows through a small pump in such a short time.” There is thus a danger that the impeller – a component inside the pump – may squash the deli
cate cells and blood platelets because it rotates 10,000 times a minute.
This must be avoided at all costs. If too many red blood cells are damaged then large quantities of the haemoglobin they contain will leak out. This substance
particles. For example, it is very time
consuming to simulate how the blood pump is started up. “This startup phase of about ten revolutions until the pump is rotating smoothly only lasts a fraction of a second. But in order to simulate this phase we need about ten seconds of computing time on the supercomputer,”
Behr explains. Behr turned to Felix Wolf for advice on how to speed up the calcu
lations on the supercomputer.
In order to distribute the calculations among a large number of processors on the supercomputer the pump is, so to speak, cut up into small pieces. One processor is then responsible for each of these segments. However, the large number of individual calculations have to be put back together again. This is where one of the difficulties lies. “What happens at the edges of the segments is often interesting,” says Felix Wolf. The exchange of data between the processors is often quite decisive here. “If problems arise then it is no good distributing the task between even more processors. The processors then simply exchange more and more messages and sometimes many of them are simply superfluous, as if they were sending empty envelopes.”
TRACKING DOWN WEAK POINTS
This is why the simulation of the flows in the DeBakey blood pump was initially no faster or more accurate when more than 900 processors were involved. Wolf used special software to get to the bot
tom of this problem. He incorporated small sensors into the simulation pro
gram. That is to say, little pieces of pro
gram that record all the details of each mathematical operation. Using the pro
gram package developed at Jülich and
known as Scalasca (Scalable Perform
ance Analysis of Large Scale Applica
tions), he analysed these data and thus tracked down the weak points in the sim
ulation program. On the basis of this analysis, it became possible to stem the flood of “empty envelopes”. “The simula
tion now runs efficiently on more than 4,000 processors and is roughly five times as fast as it was before,” Wolf is pleased to say. In this way, new versions of the pump can be examined much fast
er in a virtual test run, which means that patients also benefit more quickly. For example, a scaleddown version of the pump has been developed for children aged from five to sixteen years suffering from heart disease. “Such successes are only possible because engineers and computer scientists work hand in hand in JARA,” says Wolf.
He is hoping that the European Part
nership for Advanced Computing in Europe (PRACE) initiative will lead to fur
ther progress. Individual supercomputers are already available in Europe that are well up in the international ranking. “But we are still not in as good a position as the USA,” says Wolf. However, European plans for petaflop class computers pro
viding 1,000 trillion floating point ope
rations per second are already well advanced. “This will further improve con
ditions for researchers in Europe who want to perform complex simulations,”
Wolf confidently adds. A longterm goal may also be coming a bit closer: blood pumps which do not just keep the patient alive until a suitable donor is found but which can actually provide a permanent replacement for the human heart. ::
Wiebke Rögener
Felix Wolf between the racks of the Jülich supercomputer. Together with his team, he has made the calculations for the blood pump simulations faster and more effective.